omnigraph/docs/dev/rfc-003-mcp-server-surface.md

RFC-003: MCP Server Surface for omnigraph-server

Status: Proposed — buildable implementation spec. Date: 2026-06-13 Audience: server/engine maintainers. Tickets: MR-969 (stored queries + MCP exposure), MR-956 (OAuth/RFC-9728 layer), MR-971 (per-server credential resolver — landed as RFC-007), MR-974 (omnigraph mcp install). Builds on: rfc-001-queries-envelope-mcp.md (stored queries + the response envelope), rfc-005-server-cluster-boot.md (multi-graph boot), rfc-007-operator-config.md (client credential model), rfc-009-unify-access-paths.md (one-contract/two-implementations posture). Target release: v0.8.x.

Re-validated against main 2026-06-16 (post RFC-009 omnigraph-api-types/GraphClient, RFC-011 cluster-only server, RFC-009 canonical POST /load, RFC-012 embeddings): every file:line and Reuses citation below was re-checked against the merged tree; the deltas are folded in (cluster-only routing in §15, the DTO crate move in §9, /ingest→/load, and the per-query expose/tool_name deferral in §17). An external review pass (8 findings) was then folded in as correct-by-design fixes, not point patches — the schema generator is locked to the engine coercer by an equivalence test (§9.1), Origin is fail-closed by a single host-policy constructor (§7), the list seam is non-paginated by contract (§6), and name collisions fail at validate-time (§9.3); the resolved decisions are catalogued in §17.

Validated against (re-checked 2026-06-13): MCP protocol revision 2025-11-25 (modelcontextprotocol.io), the official Rust SDK rmcp 1.7.0 (crates.io / github.com/modelcontextprotocol/rust-sdk), and current tool/security best practice (Anthropic engineering, MCP spec security pages). Provider compatibility was checked against the live docs of Claude Code/Desktop/web, the Claude Messages API MCP connector, OpenAI's Responses API + ChatGPT connectors, Cursor, VS Code Copilot, and OpenCode. Code snippets marked Reuses are present in omnigraph-server today, cited to file:line; snippets marked New are the code to add.

---

1. Summary

Add a first-class MCP (Model Context Protocol) server surface to omnigraph-server, served over Streamable HTTP, that projects the server's operations as MCP tools and resources for LLM clients. Two tool populations share one projection path:

1. Built-in operational tools — graph read/mutate, schema get/apply, branch create/delete/merge/list, commit list/get, NDJSON load, and a graph-scoped graph_health liveness tool, plus resources omnigraph://schema and omnigraph://branches.
2. Dynamic stored-query tools — projected from the graph's loaded stored-query registry: either one typed tool per query (small catalogs) or a discovery + execute meta-tool pair (large catalogs) — see §9.

Every tool is authorized by the server's existing Cedar policy engine. The MCP layer performs no authentication of its own: it consumes an already-resolved actor identity from the server's bearer/OAuth middleware, so the same endpoint serves on-prem (static or customer-OIDC tokens) and cloud (OAuth 2.1) by configuration only. The transport is stateless JSON over a single POST endpoint — the minimal conformant Streamable-HTTP shape, since the server emits no server-initiated messages.

The surface is built so the existing local stdio MCP package can later collapse into a thin stdio↔HTTP proxy over it, leaving one Cedar-gated, remotely-reachable tool set (§13).

2. Goals

• Project built-in tools and stored queries through one registry abstraction, so tools/list / tools/call never special-case a population.
• Make tools/list and the callable set agree for argument-independent authorization, both driven by Cedar; tools/call is always the authoritative gate.
• Keep the MCP layer auth-method-agnostic: it consumes a resolved actor, never a raw token, and never branches on how authentication happened.
• Add no business logic: tools delegate to the same engine functions the HTTP routes call.
• Be code-mode-friendly (typed schemas, structured output, stable names, progressive disclosure) and maximally client-compatible (Streamable HTTP + bearer today, OAuth 2.1 + RFC 9728 as an additive layer).
• Behaviour-neutral when unused: no MCP traffic ⇒ no change.

3. Non-Goals

• Hosting an OAuth authorization server. The server is a Resource Server only (validates tokens, never issues them, never holds client secrets). The AS is a separate concern (MR-956).
• MCP prompts, elicitation, sampling, tasks, tools/list_changed subscriptions, resource subscriptions, server-initiated messages — none required, which is what permits the stateless POST-only transport. (tools/list_changed is reconsidered only if the registry gains runtime reload.)
• stdio transport inside the server — stdio stays in the TS package (later a proxy).
• Client-side "code mode" machinery (TS wrapper generation, sandboxes, tool search/deferral) — those are client/runtime concerns; see §12 for what the server does to support them and what it deliberately does not build.
• Cross-graph tool listing — per-graph catalogs only.

4. Protocol target

Target MCP revision 2025-11-25 (current). rmcp 1.7.0 advertises this as its latest and negotiates down to any of 2024-11-05 / 2025-03-26 / 2025-06-18 / 2025-11-25; an absent MCP-Protocol-Version header defaults to 2025-03-26 and an unsupported one is a 400. Revision 2025-06-18 is the floor we rely on for two features: structured tool output (outputSchema + structuredContent) and the OAuth Resource-Server model. From 2025-11-25 we adopt: input-validation errors as tool-execution errors (SEP-1303), JSON Schema 2020-12 as the default dialect, 403 on a present-but-disallowed Origin (validated fail-closed by a single host-policy constructor — §7, not a config-presence default), and WWW-Authenticate made optional with a .well-known fallback.

Transport shape (stateless Streamable HTTP). The server exposes one endpoint that accepts POST (and answers GET/DELETE with 405 + Allow: POST). For a JSON-RPC request it returns one application/json object; it opens no SSE stream, assigns no Mcp-Session-Id, and treats every request independently — a fully conformant stateless server. It MUST validate Origin (403 on mismatch) and honor MCP-Protocol-Version. rmcp delivers all of these in stateless mode (§7).

5. Crate architecture

Two crates; rmcp is contained to one of them.

omnigraph-server  (implements McpBackend; all omnigraph tool/Cedar/dispatch logic)
        │ depends on
        ▼
omnigraph-mcp     (rmcp Streamable-HTTP transport, the McpBackend trait, rmcp model re-exports)
        │ depends on
        ▼
rmcp 1.7 + tower-http(limit) + axum + http

The dependency must go server → mcp. The server binary mounts /mcp, so a mcp → server edge cycles at the package level (server-bin → omnigraph-mcp → server-lib), which Cargo rejects. The trait inverts the direction — the crate defines the seam, the server fills it — which is also why the crate can never name an omnigraph type (AppState, GraphHandle, the handlers); it abstracts over them.

crates/omnigraph-mcp/Cargo.toml:

[package]
name = "omnigraph-mcp"
edition = "2024"                 # rmcp 1.7 is itself edition 2024 — no friction
version.workspace = true

[dependencies]
# `server` is on by rmcp's default features; `transport-streamable-http-server`
# pulls in the tower service + http stack. Do NOT enable rmcp's `local` feature —
# it cfg's the StreamableHttpService tower wiring out.
rmcp = { version = "1.7", default-features = false, features = ["server", "transport-streamable-http-server"] }
axum       = { workspace = true }
http       = "1"
tower-http = { workspace = true, features = ["limit"] }
tokio      = { workspace = true }
async-trait = { workspace = true }
serde_json = { workspace = true }

Add "crates/omnigraph-mcp" to the workspace members; in omnigraph-server/Cargo.toml add omnigraph-mcp and no direct rmcp dep (verified absent today). The verification gate is cargo tree -p omnigraph-server -e normal | grep rmcp showing rmcp only transitively under omnigraph-mcp.

6. The McpBackend seam — New in omnigraph-mcp

// crates/omnigraph-mcp/src/lib.rs
use async_trait::async_trait;

// rmcp model types re-exported so the server speaks rmcp via `omnigraph_mcp::…`
// and carries no direct rmcp dependency.
pub use rmcp::model::{
    CallToolResult, Content, RawResource, ReadResourceResult, Resource,
    ResourceContents, ServerCapabilities, ServerInfo, Tool, ToolAnnotations,
};
pub use rmcp::ErrorData as McpError;        // JSON-RPC error type (method_not_found=-32601, invalid_params=-32602, internal_error=-32603)
pub type JsonObject = serde_json::Map<String, serde_json::Value>;

#[async_trait]
pub trait McpBackend: Clone + Send + Sync + 'static {
    fn server_info(&self) -> ServerInfo;
    async fn list_tools(&self, parts: &http::request::Parts) -> Result<Vec<Tool>, McpError>;
    async fn call_tool(&self, parts: &http::request::Parts, name: &str, args: JsonObject) -> Result<CallToolResult, McpError>;
    async fn list_resources(&self, parts: &http::request::Parts) -> Result<Vec<Resource>, McpError>;
    async fn read_resource(&self, parts: &http::request::Parts, uri: &str) -> Result<ReadResourceResult, McpError>;
}

The list seam is non-paginated by contract — deliberately. list_tools / list_resources return the full set, so McpService always emits nextCursor: null. This is correct-by-design for this surface, not an oversight: the catalog is bounded — built-ins are a fixed ~dozen, and a large stored-query catalog is bounded by the meta projection mode (§9.2), which collapses N queries into two tools rather than leaning on tools/list paging. The trait return type (Vec<T>) is the contract; the doc must not claim pagination the signature can't express (§12, §16 are aligned to this — no tools/list/resources/list cursor). If a future surface genuinely needs paging, that is a seam-signature change (-> ListToolsResult with a cursor), made together with the capability — never a doc promise ahead of the type.

&http::request::Parts is the decoupling mechanism. The crate hands the backend the request parts; the backend reads its own types out of parts.extensions. The crate never names an omnigraph type, so it is reusable and auth stays decoupled (§8).

rmcp's own ServerHandler trait uses RPITIT (-> impl Future + …), not async-trait. Our McpBackend deliberately uses #[async_trait]: it is implemented once by the server, the boxed future is negligible at MCP QPS, and the server already depends on async-trait. Either style compiles on edition 2024.

7. Transport — New in omnigraph-mcp

// crates/omnigraph-mcp/src/transport.rs
use std::sync::Arc;
use rmcp::transport::streamable_http_server::{
    StreamableHttpServerConfig, StreamableHttpService,
    session::never::NeverSessionManager,     // stateless ⇒ reject all session ops
};

// Host + Origin posture as a TOTAL choice — there is no `None ⇒ skip` state to leak
// into a fail-open default. `OriginPolicy` is the by-design closure for the Origin
// class: every deployment lands in exactly one arm, chosen once by `from_bind`.
pub enum OriginPolicy {
    Allow(Vec<String>),  // browser clients from these origins; any OTHER present Origin → 403
    DenyBrowsers,        // no browser clients expected; ANY present Origin → 403 (non-browser MCP clients send none)
    Unchecked,           // explicit opt-out (loopback dev / trusted network) — never the remote default
}
pub struct McpHostPolicy {
    pub allowed_hosts: Option<Vec<String>>,   // None ⇒ accept any Host (DNS-rebinding defense relaxed for a known-public bind)
    pub origin:        OriginPolicy,          // no Option — a total decision
}

impl McpHostPolicy {
    // The ONLY constructor. Host and Origin posture are derived together from the
    // bind + config, fail-closed: a remote bind with no configured origins is
    // `DenyBrowsers` (a present Origin is rejected), NOT "skip". A caller cannot
    // construct a fail-open policy because the struct has no skip-by-absence state.
    pub fn from_bind(bind: &SocketAddr, public_hosts: &[String], browser_origins: &[String]) -> Self {
        let loopback = bind.ip().is_loopback();
        Self {
            allowed_hosts: if loopback { Some(vec!["127.0.0.1".into(), "localhost".into()]) }
                           else if public_hosts.is_empty() { None } else { Some(public_hosts.to_vec()) },
            origin: if !browser_origins.is_empty() { OriginPolicy::Allow(browser_origins.to_vec()) }
                    else if loopback { OriginPolicy::Unchecked }     // local dev convenience only
                    else { OriginPolicy::DenyBrowsers },             // remote default: fail-closed
        }
    }
}

pub fn mcp_router<B: McpBackend>(backend: B, body_limit: usize, hosts: McpHostPolicy) -> axum::Router {
    // StreamableHttpServerConfig is #[non_exhaustive]; its Default is stateful_mode=true,
    // json_response=false, allowed_hosts=loopback. ALL THREE must be overridden for a
    // remote stateless JSON server — build from Default and flip via the with_* setters.
    let mut config = StreamableHttpServerConfig::default()
        .with_stateful_mode(false)
        .with_json_response(true);
    config = match &hosts.allowed_hosts {
        Some(list) => config.with_allowed_hosts(list.clone()),
        None        => config.disable_allowed_hosts(),     // accept any Host
    };
    // rmcp validates Origin ONLY when allowed_origins is non-empty (empty ⇒ rmcp skips),
    // so DenyBrowsers cannot be expressed by handing rmcp a list. We therefore enforce
    // OriginPolicy in a thin pre-layer that 403s a disallowed present Origin BEFORE rmcp
    // — making fail-closed independent of rmcp's empty-list semantics (the root cause of
    // the original fail-open default). `Allow` also configures rmcp as defense-in-depth.
    if let OriginPolicy::Allow(origins) = &hosts.origin { config = config.with_allowed_origins(origins.clone()); }

    // service_factory returns Result<S, io::Error>; NeverSessionManager pairs with stateless mode.
    let svc = StreamableHttpService::new(
        move || Ok(McpService::new(backend.clone())),
        Arc::new(NeverSessionManager::default()),
        config,
    );

    axum::Router::new()
        .route_service("/mcp", svc)
        .layer(origin_guard(hosts.origin))   // fail-closed Origin enforcement (no-op only for Unchecked)
        // rmcp reads the body directly (not via an axum extractor), so axum's
        // DefaultBodyLimit does NOT bound /mcp — the tower-http layer does.
        .layer(tower_http::limit::RequestBodyLimitLayer::new(body_limit))
}

McpService<B> implements rmcp's ServerHandler, pulls &Parts out of the request context once, and delegates each method to B. rmcp's StreamableHttpService consumes the body and injects the remaining http::request::Parts into RequestContext.extensions (this is documented and load-bearing — see §8); inside the handler, ctx.extensions.get::<http::request::Parts>() returns those parts.

Conformance the stateless transport gives for free (verified in rmcp 1.7 tower.rs): GET/DELETE /mcp → 405 with Allow: POST; a disallowed Host → 403; MCP-Protocol-Version → 400 on unsupported, default 2025-03-26 when absent. The one thing rmcp does not give for free is fail-closed Origin: rmcp checks Origin only when allowed_origins is non-empty, so an empty list is fail-open. origin_guard (above) closes that — a present, disallowed Origin → 403 regardless of rmcp's empty-list behavior. That layer is the only added middleware.

Host/Origin policy is fail-closed by construction, derived from the deployment. rmcp's default allowed_hosts is loopback-only — correct for local dev (DNS-rebinding defense) but it would 403 every remote client. McpHostPolicy::from_bind (the single constructor) computes both axes once at startup from --bind + config: loopback bind → loopback Host allow-list + OriginPolicy::Unchecked (dev convenience); non-loopback bind → the configured public host(s) (else Host-allowlisting disabled, logged — bearer is the real control), and OriginPolicy::DenyBrowsers by default (any present Origin → 403) unless browser_origins are configured (OriginPolicy::Allow). The key by-design property: OriginPolicy has no "absent ⇒ skip" state and there is no other way to build the policy, so a remote deployment cannot accidentally run fail-open — closing the bug class rather than flipping a default. Non-browser MCP clients (the Phase-1 tier) send no Origin and are unaffected; only a forged browser Origin is rejected.

8. Auth & identity — Reuses the server's middleware

The backend consumes an already-resolved actor and branches on nothing about how the token was verified. Two values are injected into the request extensions by middleware that runs before the MCP service:

// Reuses — crates/omnigraph-server/src/identity.rs:186
pub struct ResolvedActor { pub actor_id: Arc<str>, pub tenant_id: Option<TenantId>, pub scopes: Vec<Scope>, pub source: AuthSource }

// Reuses — crates/omnigraph-server/src/registry.rs:37
pub struct GraphHandle {
    pub key: GraphKey, pub uri: String,
    pub engine: Arc<Omnigraph>,
    pub policy:  Option<Arc<PolicyEngine>>,   // None ⇒ no per-graph Cedar gate
    pub queries: Option<Arc<QueryRegistry>>,  // None ⇒ no stored queries for this graph
}

The middleware order is fixed in build_app (lib.rs:876; the two route_layers at lib.rs:929-936): the outer layer require_bearer_auth injects Extension<ResolvedActor> (or 401); the inner layer resolve_graph_handle injects Extension<Arc<GraphHandle>>. Both land in request.extensions(), which rmcp copies into RequestContext.extensions.

// New — crates/omnigraph-server/src/mcp/mod.rs
#[derive(Clone)]
pub struct OmnigraphMcpBackend { state: AppState }   // AppState is Arc-backed #[derive(Clone)]

impl OmnigraphMcpBackend {
    fn ctx<'a>(&self, parts: &'a http::request::Parts) -> Result<(&'a ResolvedActor, &'a Arc<GraphHandle>), McpError> {
        let actor  = parts.extensions.get::<ResolvedActor>()
            .ok_or_else(|| McpError::internal_error("actor missing from request extensions", None))?;
        let handle = parts.extensions.get::<Arc<GraphHandle>>()
            .ok_or_else(|| McpError::internal_error("graph handle missing from request extensions", None))?;
        Ok((actor, handle))
    }
}

Auth posture (spec-aligned, MCP 2025-11-25 authorization). The server is a Resource Server. Per-request validation only — sessions are never used for authentication (the transport is stateless, which makes this structural). Token audience must be validated and token passthrough is prohibited: if a tool later needs to reach an upstream API, the server acts as a separate OAuth client and must not forward the client's token.

• Static bearer (today). require_bearer_auth resolves a ResolvedActor from a SHA-256 hash match. Works for the developer/agent clients (§13).
• OAuth 2.1 + RFC 9728 (additive, MR-956). Serve /.well-known/oauth-protected-resource; on 401, optionally add WWW-Authenticate: Bearer resource_metadata="…" (header is optional in 2025-11-25 given the well-known fallback). Clients run OAuth 2.1 + PKCE + RFC 8707 resource indicators themselves; the server validates audience-bound JWTs offline (cached JWKS), so on-prem/air-gapped keeps working. This swaps the bearer middleware behind a TokenVerifier and changes zero MCP code.

Compatibility caveat to honor (Claude Code issue #59467). Advertising RFC-9728 Protected-Resource-Metadata can cause some clients (Claude Code today) to ignore a static Authorization header and force the OAuth flow. So PRM advertisement must be config-gated: a deployment serving developer clients over static bearer does not advertise OAuth; a deployment targeting consumer connectors does. The MCP routes only need to flow through the standard 401 path so the hook can be added without touching MCP code.

9. Stored-query projection

The projection source is the same query_catalog_entry the GET /queries catalog uses (crates/omnigraph-server/src/api.rs:13). The param/catalog DTOs moved to the shared omnigraph-api-types crate (RFC-009 Phase 2) and are re-exported through api.rs (pub use omnigraph_api_types::*), so the Reuses types below still resolve via omnigraph_server::api::…. Real types:

// Reuses — crates/omnigraph-api-types/src/lib.rs:355 (re-exported via omnigraph-server/src/api.rs)
pub enum ParamKind { String, Bool, Int, BigInt, Float, Date, DateTime, Blob, Vector, List }

// Reuses — crates/omnigraph-api-types/src/lib.rs:373
pub struct ParamDescriptor {
    pub name: String,
    pub kind: ParamKind,
    pub item_kind: Option<ParamKind>,   // Some(scalar) when kind == List
    pub vector_dim: Option<u32>,        // Some(dim) when kind == Vector — the dimension lives here, not in the kind
    pub nullable: bool,
}

// Reuses — crates/omnigraph-server/src/queries.rs:29
pub struct StoredQuery { pub name: String, pub source: Arc<str>, pub decl: QueryDecl, pub expose: bool, pub tool_name: Option<String> }
impl StoredQuery { pub fn is_mutation(&self) -> bool; pub fn effective_tool_name(&self) -> &str; }   // queries.rs:45,55
pub struct QueryRegistry { /* by_name: BTreeMap<String, StoredQuery>; .lookup(&name) */ }            // queries.rs:64

A query is declared in the cluster's cluster.yaml graphs.<id>.queries (a directory to discover, an explicit file list, or a name: { file: … } map); cluster apply publishes it to the content-addressed catalog, and the server loads that graph's applied registry into handle.queries at boot (settings.rs:71-111). The StoredQuery struct carries expose: bool and tool_name: Option<String>, but cluster boot currently forces expose: true, tool_name: None for every applied query (settings.rs:83-84, the §D5 bridge — see §17). So today the projection lists every applied query and names each by its query name; the expose/tool_name plumbing is wired but inert until the cluster catalog grows the per-query metadata. The projection reads handle.queries and is agnostic to the declaration source. (The legacy single-graph omnigraph.yaml queries: map is removed — RFC-011 made the server cluster-only; there is no other declaration source.)

9.1 ParamDescriptor → JSON Schema (New, shared projection + equivalence test)

JSON Schema 2020-12. The schema generator is the engine's input contract, not a second copy of it. The authority for what a param accepts is the runtime coercer json_value_to_literal_typed (crates/omnigraph-compiler/src/query_input.rs); a hand-written schema in the MCP crate is a parallel encoding that will drift — the review found two drifts at once (Blob, nullable), and BigInt/Date/Vector are latent siblings of the same class. So the projection lives next to the DTO it projects, in omnigraph-api-types (where ParamKind/ParamDescriptor already live and are ToSchema), is the single mapping both OpenAPI and MCP consume, and is locked to the coercer by an equivalence test — drift becomes a CI failure, not a shipped bug.

// New — crates/omnigraph-api-types/src/lib.rs (next to ParamKind/ParamDescriptor)
use serde_json::{json, Value};

// Exhaustive, wildcard-free: adding a ParamKind is a COMPILE error until its arm
// (and its equivalence-test corpus row) exist — closing "new kind, wrong/default schema".
fn scalar_schema(kind: ParamKind) -> Value {
    match kind {
        ParamKind::String   => json!({ "type": "string" }),
        ParamKind::Bool     => json!({ "type": "boolean" }),
        ParamKind::Int      => json!({ "type": "integer" }),
        ParamKind::BigInt   => json!({ "type": "string", "pattern": r"^-?\d+$" }), // i64/u64 lose precision >2^53 as JSON numbers
        ParamKind::Float    => json!({ "type": "number" }),
        ParamKind::Date     => json!({ "type": "string", "format": "date" }),
        ParamKind::DateTime => json!({ "type": "string", "format": "date-time" }),
        // FIX (③): the coercer takes Blob as a blob-URI STRING ("expected blob URI
        // string", query_input.rs:449; DTO doc api-types:354) — NOT base64-decoded bytes.
        ParamKind::Blob     => json!({ "type": "string", "format": "uri" }),
        ParamKind::Vector | ParamKind::List => unreachable!("composite kinds handled in param_json_schema"),
    }
}

// The one entry point the MCP crate calls — applies the nullable rule uniformly.
pub fn param_json_schema(p: &ParamDescriptor) -> Value {
    let base = match p.kind {
        ParamKind::Vector => {
            let mut s = json!({ "type": "array", "items": { "type": "number" } });
            if let Some(dim) = p.vector_dim { s["minItems"] = json!(dim); s["maxItems"] = json!(dim); }
            s
        }
        ParamKind::List => json!({ "type": "array", "items": p.item_kind.map(scalar_schema).unwrap_or_else(|| json!({"type":"string"})) }),
        scalar => scalar_schema(scalar),
    };
    // FIX (④): the coercer accepts explicit `null` for a nullable param AND its
    // omission (query_input.rs:273,296). `required` alone only covers omission; a
    // strictly-validating client (or SEP-1303 input validation) would reject `null`
    // against the bare scalar. Allow null at the schema level for nullable params.
    if p.nullable { json!({ "anyOf": [ base, { "type": "null" } ] }) } else { base }
}

The lock — an equivalence test (the by-design closure), in the compiler crate (it sees both the coercer and param_json_schema): for a fixed accept/reject corpus per ParamKind (incl. a blob-URI string, a base64 blob that must now validate as a plain string, null for nullable vs non-nullable, an over/under-length vector), assert schema_accepts(v) == json_value_to_literal_typed(name, v, kind, mode).is_ok(). Any future arm that diverges from the engine — base64 creeping back, a missing null-union, a new kind without a schema — turns the test red. That test, not reviewer vigilance, is what makes the schema correct by construction.

9.2 Two projection modes (small vs large catalogs)

Tool-overload is real: model accuracy degrades sharply as a single client's tool count climbs past a few dozen, and clients that don't defer tool loading (e.g. OpenCode) pay the full tools/list token cost. So the projection has two modes, selected per graph by a stored_query_mode setting (default auto).

Where the setting lives (by-design, ⑥). There is no free-floating mcp.* key. stored_query_mode and its threshold belong to the same per-graph mcp: metadata block that will hold expose/tool_name (the cluster Phase-6 surface, §D5 bridge — see §17) — one mcp-config home, one validator, validated at cluster validate/boot with the rest of the registry. That sequences it correctly: the knob cannot land before the surface that holds it exists, and it can't drift into a second config location. Until Phase 6, the mode is not configurable — every graph runs auto (the count-based default below), which is the safe, documented behavior. The modes themselves:

• per_query (small/stable catalogs). One tool per expose: true query, named by effective_tool_name(), with a fully typed input_schema. This is the richest surface — each query is a first-class typed tool, ideal for code-mode runtimes that compile tools into a typed API.
• meta (large/dynamic catalogs). Two tools instead of N: stored_query_list(filter?, detail_level?) (returns names + descriptions; full param schema only at higher detail) and stored_query_run(name, params, branch?, snapshot?). This keeps tools/list small and mirrors the progressive-disclosure shape (search + execute) that scales to hundreds of queries.
• auto picks per_query below a threshold (default 24 exposed queries) and meta at or above it; the threshold is configurable. The boundary and count are logged so a deployment never silently flips modes.

9.3 Envelope (collision-free by construction)

In per_query mode the tool's input_schema nests query params under params, mirroring POST /queries/{name}:

{ "type": "object",
  "properties": {
    "params":   { "type": "object", "properties": { /* per-param param_json_schema(...) */ }, "required": [ /* names where nullable == false */ ] },
    "branch":   { "type": "string" },
    "snapshot": { "type": "string" }      // omit for mutation tools — mutation-against-snapshot is unrepresentable
  },
  "additionalProperties": false }

required lists only non-nullable param names; a nullable param is both absent from required and carries the null-union from param_json_schema (§9.1), so omitting it and passing explicit null both validate — matching the coercer.

Knobs (branch/snapshot) and the query's own params live in separate namespaces, so a query parameter literally named branch/snapshot cannot collide.

Built-in vs stored name collision is a load-time error, never a silent skip (⑦). The earlier "a colliding stored tool is skipped (built-ins win)" is a silent failure — a query an operator published just vanishes from the catalog at projection time, which the deny-list in docs/dev/invariants.md forbids. By-design fix: fold the built-in tool names (a stable closed set from the Builtin enum, §10) into the same per-graph uniqueness check the registry already runs at load (duplicate_tool_name, today stored-vs-stored only). A stored effective_tool_name() that shadows a built-in then fails cluster validate/server boot loudly, before serving — a runtime-shadowed query becomes structurally impossible rather than silently dropped.

10. Tool catalog + Cedar mapping — Reuses PolicyAction

Each built-in reuses the exact PolicyAction its REST route enforces:

// Reuses — crates/omnigraph-policy/src/lib.rs:16
pub enum PolicyAction {
    Read, Export, Change, SchemaApply,
    BranchCreate, BranchDelete, BranchMerge,
    Admin,        // reserved, no call site yet
    GraphList,    // server-scoped (resource_kind == Server)
    InvokeQuery,  // graph-scoped, coarse (no per-query dimension yet)
}

A tool's scope is derived from where it is mounted, not asserted independently: MCP mounts only under /graphs/{graph_id}/mcp (§15), so every MCP tool is graph-scoped by construction. There is no server-scoped MCP tool — a "server-scoped tool on a per-graph mount" is unrepresentable (⑧). Server-level liveness stays on REST GET /healthz; the MCP liveness tool is graph-scoped graph_health (confirms this graph's handle is live) and needs no Cedar gate.

MCP tool	Scope	Cedar action
graph_health	graph	none (liveness/version)
graph_snapshot, schema_get, branch_list, commit_list, commit_get	graph	Read
graph_query (ad-hoc read)	graph	Read (run_query self-authorizes)
graph_mutate (ad-hoc write)	graph	Change
graph_load (NDJSON)	graph	Change (+ BranchCreate iff from is present — see §11)
branch_create / branch_delete / branch_merge	graph	BranchCreate / BranchDelete / BranchMerge
schema_apply (allow_data_loss)	graph	SchemaApply
stored query (per_query) / stored_query_run (meta)	graph	InvokeQuery (coarse) then inner Read/Change

Naming. Tool ids are domain-qualified snake_case (graph_query, branch_merge, schema_apply, …) within the spec's [A-Za-z0-9_.-], 1–128-char constraint. Domain qualification (rather than bare query/mutate) reduces cross-server collisions when a client loads omnigraph alongside other MCP servers; clients that auto-prefix by connection name (e.g. OpenCode → omnigraph_graph_query) compose cleanly. Names are a stability contract (Hyrum's Law) — don't churn them.

Annotations (set explicitly). MCP annotation defaults are pessimistic (readOnlyHint=false, destructiveHint=true, idempotentHint=false, openWorldHint=true), so an unannotated read tool is mistaken for a destructive open-world writer. Set them via rmcp's ToolAnnotations (read_only_hint, destructive_hint, idempotent_hint, open_world_hint):

• read tools (graph_query, graph_snapshot, schema_get, branch_list, commit_*, stored reads) → read_only_hint = true, open_world_hint = false.
• writers (graph_mutate, graph_load, branch_delete, branch_merge, schema_apply) → read_only_hint = false, destructive_hint = true, open_world_hint = false. Clients use destructiveHint to drive human-confirmation prompts.
• branch_create (additive) → destructive_hint = false.

Annotations are advisory hints, not a security boundary (clients may ignore them); Cedar is the enforcement boundary.

Represent built-ins as a Builtin enum (one variant per tool; descriptor / gate / call as match arms) — lower liability than ~13 unit structs + dyn. Stored-query tools are a sibling populator over handle.queries.

list_tools / list_resources are Cedar-filtered by running the same authorization the call path runs, with default args (branch main) — not a branch: None probe (which matches no branch_scope rule and would hide tools the actor can call on a scoped branch). Over-showing a branch-scoped grant is the safe direction; call_tool is the authoritative gate.

11. Dispatch reuse + error classification

call_tool adds no business logic. Reuse points (all in handlers.rs):

pub(crate) enum Authz { Allowed, Denied(String) }                                  // handlers.rs:313
pub(crate) fn authorize(actor: Option<&ResolvedActor>, policy: Option<&PolicyEngine>, request: PolicyRequest) -> Result<Authz, ApiError>; // :334 — Err = operational 401/500
pub(crate) async fn run_query(handle: Arc<GraphHandle>, actor: Option<&ResolvedActor>, query: &str, name: Option<&str>, params_json: Option<&Value>, branch: Option<String>, snapshot: Option<String>, reject_mutations: bool) -> Result<(String, ReadTarget, QueryResult), ApiError>; // :711
pub(crate) async fn run_mutate(state: AppState, handle: Arc<GraphHandle>, actor: Option<&ResolvedActor>, query: &str, name: Option<&str>, params_json: Option<&Value>, branch: String) -> Result<ChangeOutput, ApiError>; // :645

PolicyRequest carries { action, branch, target_branch } only — no actor identity (server-resolved, supplied separately) and no query-name dimension (the coarse-invoke_query caveat):

// Reuses — crates/omnigraph-policy/src/lib.rs:251
pub struct PolicyRequest { pub action: PolicyAction, pub branch: Option<String>, pub target_branch: Option<String> }

The stored-query double-gate + deny-masking pattern (handlers.rs:913, server_invoke_query) is the contract call_tool mirrors for stored queries:

// Reuses (pattern) — outer InvokeQuery gate; deny == missing so the catalog can't be probed
match authorize(actor, handle.policy.as_deref(), PolicyRequest {
    action: PolicyAction::InvokeQuery, branch: None, target_branch: None,   // graph-scoped: NO branch dimension
})? {
    Authz::Allowed => {}
    Authz::Denied(_) => return Err(ApiError::not_found("stored query not found")),
}
let stored = handle.queries.as_ref().and_then(|r| r.lookup(&name)).ok_or_else(|| ApiError::not_found("stored query not found"))?;
// inner gate runs in run_mutate (Change) / run_query (Read); a stored mutation is double-gated.

graph_load (NDJSON) wraps the unified load_as via run_ingest (the canonical server_load handler, handlers.rs:1320; POST /ingest / server_ingest, handlers.rs:1360, is a #[deprecated] alias emitting RFC-9745 headers — RFC-009 Phase 5): a missing branch with no from is a 404, never an implicit fork; BranchCreate is consulted only when from is present, then Change for the load. The tool's input_schema is { data: string, branch?: string, from?: string, mode?: "merge"|"append"|"overwrite" }, additionalProperties: false (the same IngestRequest shape, omnigraph-api-types/src/lib.rs:496).

Error classification (New, one mapper, SEP-1303-aligned). ApiError's fields are private (lib.rs:280, and still carry no public status/message accessors), so add pub(crate) fn status_code(&self)/message_str(&self) accessors. Then one classify is used at every dispatch site:

// New — the single source of truth
fn classify(r: Result<CallToolResult, ApiError>) -> Result<CallToolResult, McpError> {
    match r {
        Ok(out) => Ok(out),
        // Semantic failures (bad params, validation, business 4xx/409) → isError result,
        // fed back to the model so it self-corrects (MCP 2025-11-25 SEP-1303).
        Err(e) if e.status_code().is_client_error() => Ok(CallToolResult::error(vec![Content::text(e.message_str())])),
        // Operational failures (5xx) → JSON-RPC protocol error.
        Err(e) => Err(McpError::internal_error(e.message_str().to_owned(), None)),
    }
}

Two cases are protocol errors, not isError, so the catalog isn't probeable and malformed calls are unambiguous: an unknown OR denied tool returns an identical McpError::invalid_params("unknown tool: <name>") (-32602), and a structurally malformed call (failing the tools/call shape) is a protocol error. A missing/bad bearer is an HTTP 401 at the boundary, before rmcp.

12. Code-mode compatibility

"Code mode" (Anthropic's Code execution with MCP; Cloudflare's Code Mode) is a client/runtime technique: the client compiles a server's tools into a typed code API (TS modules / a sandbox), the model writes code against it, and intermediate results are filtered in the sandbox instead of round-tripping through the model context (reported ~98% context savings on large workflows). It runs over standard tools/list + tools/call and requires no new server endpoints; credentials stay in the transport and the runtime holds them (the sandbox never sees the bearer).

The server's job is to be a good source for that compilation. Concrete server-side choices this RFC adopts:

1. Strict, fully-typed input_schema (§9.1) with additionalProperties: false, enums for mode/format, explicit required — these compile into precise TS input types.
2. Structured output — see §13.1: declare outputSchema and return structuredContent so generated code gets typed returns, not any.
3. Stable, descriptive tool names + rich descriptions (§10) — names become function names; descriptions become doc comments.
4. Progressive disclosure for large catalogs — the meta projection mode (§9.2) keeps tools/list small (stored_query_list + stored_query_run), the same search + execute shape code-mode runtimes prefer.
5. Bounded tools/list instead of pagination. The list seam is non-paginated by contract (§6); a large catalog is bounded by the meta mode (§9.2), not by cursor paging. This keeps the seam type honest (no nextCursor the Vec<T> return can't carry) while still preventing context blow-up on big query catalogs.
6. Schemas as resources (§14) — expose the graph schema (and per-query param schemas) as MCP resources, the on-demand channel code-mode clients pull from.
7. Auth in the transport only — never require secrets as tool arguments (that would put them in model context / generated code and break the sandbox's credential isolation).

The server deliberately does not build TS-wrapper generation, sandboxes, tool search/deferral, or PII tokenization — those are client/runtime concerns, and there is no ratified "tools-as-code" MCP spec to target.

13. Provider compatibility

Transport: Streamable HTTP is the universal target — every current client below supports it for remote servers, and it is the recommended transport over deprecated HTTP+SSE.

Auth splits the ecosystem into two tiers:

Client	Remote transport	Auth that works	Notes
Claude Code (CLI)	Streamable HTTP	static bearer header and OAuth 2.1	claude mcp add --transport http <url>/mcp --header "Authorization: Bearer …". Advertising RFC-9728 can override the static header (issue #59467) — gate PRM.
Cursor	Streamable HTTP	static header and OAuth 2.1	"headers": {"Authorization": "Bearer ${env:…}"} in mcp.json.
VS Code (Copilot agent)	Streamable HTTP	static header and OAuth	needs VS Code ≥ 1.101 for remote + OAuth; auto-detects 401 → sign-in.
OpenCode	remote HTTP	static header and OAuth (auto, DCR)	mcp block in opencode.json; auto-prefixes tools omnigraph_…; no progressive disclosure → keep the static surface tight (favors meta mode at scale).
Claude Messages API (mcp_servers)	Streamable HTTP (+SSE)	pre-acquired token via authorization_token	forwards a token; never runs OAuth. Static bearer fits directly. Pin the beta header you target.
OpenAI Responses API (mcp tool)	Streamable HTTP (+SSE)	pre-acquired token via the dedicated authorization field	forwards the token on Authorization (static bearer fits directly); never runs OAuth. require_approval gates tool calls. (Current docs expose authorization, not a free-form headers object — ⑤.)
ChatGPT (developer mode/connectors)	Streamable HTTP (+SSE)	OAuth, No-Auth, or Mixed	beta; OAuth is the clean path.
Claude Desktop (custom connectors)	Streamable HTTP (+SSE)	OAuth 2.1 or authless	no static-header field — bearer-only deployments are unreachable without a gateway.
Claude.ai web (custom connectors)	Streamable HTTP (+SSE)	OAuth 2.1 + RFC 9728 (or authless)	server must serve RFC-9728 PRM; no static-header field.

Phased auth recommendation:

• Phase 1 — static bearer (this RFC). Reaches Claude Code, Cursor, VS Code Copilot, OpenCode, the Claude Messages API connector, and the OpenAI Responses API — the entire developer/agent/API tier. This is the correct launch posture.
• Phase 2 — OAuth 2.1 + RFC 9728 (MR-956, additive). Required to reach claude.ai web and Claude Desktop custom connectors and the clean ChatGPT path. The same endpoint accepts validated OAuth access tokens and (still) static bearers; PRM advertisement stays config-gated because of the #59467 header-override behavior.

Because the resource server validates whatever token arrives on Authorization, both tiers hit one endpoint with no MCP-layer branching.

13.1 Result shaping & structured output

For typed, machine-consumable results (graph_query, stored-query reads, branch_list, commit_*, schema_get) the tool declares an outputSchema and returns structuredContent (the route's existing ReadOutput / listing DTOs, which already derive ToSchema), and also mirrors the JSON in a text Content block for clients that don't parse structured content. Plain text-JSON is used where a fixed schema is awkward. (Some clients still mishandle structuredContent: null — emit an empty object, never null, when there is no structured payload.)

14. Resources

Two resources: omnigraph://schema (Read → schema .pg text) and omnigraph://branches (Read → branches JSON). Both are Cedar-filtered and deny-masked exactly like tools — a locked-down agent denied Read never sees them, which is how the "agents don't introspect schema" intent is met by policy, not omission. Advertise the resources capability with subscribe:false, listChanged:false (both handlers are backed — don't advertise a capability whose read would 404). Exposing the schema as a resource is also the on-demand channel code-mode clients pull from (§12).

No omnigraph://graphs resource and no graphs_list tool — server-scoped graph discovery stays REST-only via GET /graphs (§15).

15. Routing — Reuses build_app

/mcp is merged into per_graph_protected, which build_app always nests under /graphs/{graph_id}. RFC-011 made the server cluster-only — there is no flat single-graph route group and no match state.routing(), so /mcp is always /graphs/{graph_id}/mcp (even a single-graph boot builds a one-graph registry keyed by default; GraphRouting is now { registry, config_path }):

// Reuses — crates/omnigraph-server/src/lib.rs:876 (abridged)
let per_graph_protected = Router::new()
    .route("/snapshot", get(server_snapshot))
    // … /query /mutate /queries /queries/{name} /schema /schema/apply /load /branches /commits …
    .merge(mcp::mcp_router(state.clone()))                       // ← ADD: brings its own tower-http body-limit layer
    .route_layer(middleware::from_fn_with_state(state.clone(), resolve_graph_handle))  // inner: injects Arc<GraphHandle> (lib.rs:929)
    .route_layer(middleware::from_fn_with_state(state.clone(), require_bearer_auth));  // outer: injects ResolvedActor / 401 (lib.rs:933)

let management = Router::new()
    .route("/graphs", get(server_graphs_list))                  // GraphList — server-scoped, REST-only
    .route_layer(middleware::from_fn_with_state(state.clone(), require_bearer_auth));

// RFC-011 cluster-only: per-graph routes ALWAYS nest under /graphs/{graph_id};
// there is no flat mode and no routing match. (lib.rs:953)
let protected = Router::new()
    .nest("/graphs/{graph_id}", per_graph_protected)            // → POST /graphs/{id}/mcp
    .merge(management);

mcp::mcp_router(state) is the server's thin wrapper: omnigraph_mcp::mcp_router(OmnigraphMcpBackend::new(state), INGEST_REQUEST_BODY_LIMIT_BYTES /* lib.rs:148, 32 MiB */, host_policy_from_bind(…)). Merging the router (rather than .route("/mcp", …)) keeps the /mcp-specific body limit from leaking onto the other routes.

No server-scoped MCP. Every MCP tool/resource is graph-scoped. tools/list can't carry a graph id, so a single flat /mcp taking graph_id per call couldn't list per-graph stored-query tools and would break isolation — hence per-graph routing. A future server-level flat /mcp (bearer-only, no handle, server-scoped tools only) would live in the management group, but is not built speculatively.

15.1 Multi-graph model

omnigraph's MCP is per-graph: one isolated MCP server per graph, with the graph identity in the URL path, never in tool arguments or output. The server is cluster-only (RFC-011), so the router always nests the whole protected group under /graphs/{graph_id} (lib.rs:954) — this per-graph model is now the only model, not a multi-mode special case. Each /graphs/{id}/mcp endpoint's initialize / tools/list / tools/call / resources/* operate only on that graph and can never list or touch another graph's tools.

• Discovery is REST-only, not an MCP tool. graphs_list / omnigraph://graphs are deliberately absent from MCP. Which graphs exist is answered by GET /graphs (multi-mode only) → GraphListResponse { graphs: [{ graph_id, uri }] } (api.rs:703), gated by the server-scoped GraphList Cedar action and default-denied without a server policy (the registry — graph ids + storage URIs — is never leaked until an operator authorizes it). An operator discovers graphs via REST, then points each MCP client connection at the relevant /graphs/{id}/mcp; no single MCP connection ever sees the full graph list.

• Clients configure one connection per graph. Tool ids are identical across graphs (each is its own server), so the connection name is the namespace: a client that auto-prefixes yields og-sales_graph_query vs og-hr_graph_query.

  claude mcp add og-sales --transport http https://host/graphs/sales/mcp --header "Authorization: Bearer …"
  claude mcp add og-hr    --transport http https://host/graphs/hr/mcp    --header "Authorization: Bearer …"
  

• Stored queries are per-graph state. Each graph owns its registry (GraphHandle.queries, registry.rs:55), loaded from that graph's declaration (cluster.yaml graphs.<id>.queries). So a query is exposed only on its own graph's endpoint; the same query name may exist on multiple graphs with different definitions (no cross-graph collision — different servers). effective_tool_name() uniqueness is enforced per graph at registry load (duplicate_tool_name), not across graphs. The projection mode (per_query vs meta, §9.2) is chosen from that graph's exposed-query count, so a small graph can show one typed tool per query while a large graph on the same server uses the stored_query_list + stored_query_run meta pair. InvokeQuery is evaluated against that graph's handle.policy, so an actor can be allowed stored queries on one graph and denied on another, independently. The per-graph catalog is also discoverable over REST at GET /graphs/{id}/queries.

So tools/list on /graphs/sales/mcp returns sales' built-ins + sales' stored queries; the same call on /graphs/hr/mcp returns hr's — two disjoint catalogs, each Cedar-filtered to the actor.

16. Tests & verification

MCP tests land in a new crates/omnigraph-server/tests/mcp.rs suite (black-box over build_app); stored-query projection tests extend stored_queries.rs.

• Protocol: initialize + advertised {tools, resources} caps; tools/list returns the full bounded set with no nextCursor (the non-paginated contract, §6); tools/call happy path; GET /mcp → 405; MCP-Protocol-Version 400/default; unknown/denied tool → identical -32602.
• Origin (fail-closed, ①): remote bind, no configured origins → a present Origin is 403 (DenyBrowsers); absent Origin → 200 (non-browser clients); a configured-allowed Origin → 200; a present non-allowed Origin under OriginPolicy::Allow → 403. Asserts origin_guard, not rmcp's empty-list path.
• Cedar: a read-only actor sees read tools but not writers; a denied call masks byte-identically to an unknown one; stored queries appear only with invoke_query; the double-gate (an invoke_query-only actor sees a stored tool but the call surfaces isError when the inner Read denies).
• Dispatch: a graph_mutate writes end-to-end (proves the actor/handle extension passthrough); a malformed query → isError:true, not a protocol error; graph_load with a missing branch and no from → isError (404), with from → forks.
• Schema/engine equivalence (the by-design lock, ③④): the corpus test in the compiler crate asserting param_json_schema accepts exactly what json_value_to_literal_typed accepts, per ParamKind — incl. Blob as a URI string (a base64 blob validates only as a plain string, never decoded), explicit null for a nullable param vs rejection for a non-nullable one, list items, and vector with and without vector_dim (absent-dim omits minItems/maxItems). A drifted arm turns this red.
• Tool-name collision (⑦): a stored query whose effective_tool_name() equals a built-in fails cluster validate/boot with a loud error — it is never silently skipped or served.
• Structured output: outputSchema present and structuredContent validates against it; the text mirror is present; never emits structuredContent: null.
• Projection modes: per_query below the threshold, meta at/above it, with the switch logged.
• Auth decoupling: /mcp 401s without a bearer (before rmcp) and 200s with one; green under the static-hash verifier and a mock OIDC ResolvedActor.
• Crate-level: omnigraph-mcp/tests/ with a trivial McpBackend proving the crate stands alone (initialize + GET → 405), plus an rmcp surface guard pinning StreamableHttpServerConfig's with_* setters, NeverSessionManager, the ServerHandler method shapes, and the RequestContext.extensions → http::request::Parts passthrough — the smoke check on any rmcp bump.

Verification commands:

cargo build --workspace --locked
cargo tree -p omnigraph-server -e normal | grep rmcp     # rmcp only transitively under omnigraph-mcp
cargo test -p omnigraph-server --test mcp
cargo test -p omnigraph-server --test stored_queries
cargo test -p omnigraph-server --test openapi            # /mcp carries no #[utoipa::path]; no REST drift

17. Decisions & rollout

Locked: rmcp 1.7 (official SDK); MCP target 2025-11-25; stateless JSON over a single /mcp POST (NeverSessionManager, stateful_mode=false, json_response=true); McpBackend crate-trait seam with &Parts passthrough; Builtin enum + stored-query populator; domain-qualified snake_case tool ids; annotations set explicitly; coarse InvokeQuery with the double-gate; per-graph /mcp routing, no server-scoped MCP; structuredContent + outputSchema (with a text mirror) for typed results; vector_dim: Option<u32> handled with omit-on-absent; auth consumed as a resolved actor, validated per-request, never passed through.

Locked (correct-by-design fixes from the external review pass): one shared param_json_schema in omnigraph-api-types (Blob → URI string, nullable → null-union) co-located with the coercer and pinned by a schema/engine equivalence test — schema drift is a CI failure, not a shipped bug (③④); a non-paginated list seam by contract — meta mode bounds large catalogs, the seam type carries no nextCursor it can't honor (②); a single fail-closed McpHostPolicy::from_bind with a total OriginPolicy (no absent-⇒-skip state; remote default DenyBrowsers enforced by origin_guard) (①); built-in/stored name collisions rejected at cluster validate/boot, never silently skipped (⑦); stored_query_mode folded into the one per-graph mcp: block (Phase 6), not a floating key (⑥); MCP scope derived from the per-graph mount, so graph_health replaces a server-scoped health (⑧). The OpenAI row is corrected to the authorization field (⑤, doc-only).

Open / deferred:

• OAuth 2.1 + RFC 9728 (MR-956) — additive Phase 2; PRM advertisement config-gated (issue #59467).
• Per-query expose / tool_name (cluster Phase 6, the §D5 bridge) — the StoredQuery fields exist and the projection already reads them, but cluster boot forces expose: true, tool_name: None for every applied query (settings.rs:83-84), so today every applied query is listed and named by its query name. Per-query exposure/naming controls (mcp.expose, tool_name) land when the cluster catalog grows the metadata — no projection change is then needed.
• Per-query invoke_query scope (PR 0b) — add a query-name dimension to PolicyRequest + the Cedar schema so an actor can be scoped to specific stored queries. Until then curation is graph-level (registry membership; expose once Phase 6 lands).
• tools/list_changed — only if the registry gains runtime reload.
• stdio → proxy collapse — the local stdio package degrades to a stdio↔HTTP proxy over /mcp once this surface is GA, leaving one tool set and one Cedar gate, the same one-contract posture as rfc-009-unify-access-paths.md.

Rollout: (1) the omnigraph-mcp crate + transport + a trivial backend (crate stands alone); (2) the server backend — extension passthrough, Builtin enum, read-only tools + resources, Cedar-filtered listing, the classify mapper; (3) mutating tools + stored-query projection (both modes) + structured output; (4) docs + the omnigraph mcp install on-ramp (MR-974); (5) OAuth/RFC-9728 (MR-956) and the stdio proxy as separate follow-ups.