feat: refactor node spec and add mcp tools (#244)

* refactor: carve out extraction panel

* refactor: create spec versions for node types

* refactor: create a GenericNode and remove custom nodes

* feat: add python and typescript sdk

* add dograh sdk

* fix: fetch draft workflow definition over published one

* fix: fix routes of SDKs to use code gen

* chore: remove doclink dependency to reduce image size

* chore: format files

* chore: bump pipecat

* feat: let mcp fetch archived workflows on demand

* chore: fix tests

* feat: add sdk documentation

* chore: change banner and add badge

api/services/workflow/qa/analysis.py

@@ -8,6 +8,7 @@ from loguru import logger
 from api.db.models import WorkflowRunModel
 from api.services.gen_ai.json_parser import parse_llm_json
 from api.services.pipecat.service_factory import create_llm_service_from_provider
+from api.services.workflow.dto import QANodeData
 from api.services.workflow.qa.conversation import (
     build_conversation_structure,
     format_transcript,
@@ -77,7 +78,7 @@ async def _generate_conversation_summary(
 
 
 async def run_per_node_qa_analysis(
-    qa_node_data: dict[str, Any],
+    qa_data: QANodeData,
     workflow_run: WorkflowRunModel,
     workflow_run_id: int,
     workflow_definition: dict,
@@ -106,18 +107,16 @@ async def run_per_node_qa_analysis(
         logger.info(
             f"Events lack node_id for run {workflow_run_id}, falling back to whole-call QA"
         )
-        return await _run_whole_call_qa_analysis(
-            qa_node_data, workflow_run, workflow_run_id
-        )
+        return await _run_whole_call_qa_analysis(qa_data, workflow_run, workflow_run_id)
 
-    system_prompt = qa_node_data.get("qa_system_prompt", "")
+    system_prompt = qa_data.qa_system_prompt or ""
     if not system_prompt:
         logger.warning("No system prompt defined for QA Node")
         return {"error": "no_system_prompt", "node_results": {}}
 
     # Resolve LLM config
     provider, model, api_key, service_kwargs = await resolve_llm_config(
-        qa_node_data, workflow_run
+        qa_data, workflow_run
     )
     if not api_key:
         logger.warning(
@@ -127,7 +126,7 @@ async def run_per_node_qa_analysis(
 
     # Ensure node summaries
     node_summaries = await ensure_node_summaries(
-        workflow_definition, definition_id, workflow_run, qa_node_data
+        workflow_definition, definition_id, workflow_run, qa_data
     )
 
     # Set up Langfuse tracing
@@ -228,7 +227,7 @@ async def run_per_node_qa_analysis(
 
 
 async def _run_whole_call_qa_analysis(
-    qa_node_data: dict[str, Any],
+    qa_data: QANodeData,
     workflow_run: WorkflowRunModel,
     workflow_run_id: int,
 ) -> dict[str, Any]:
@@ -254,13 +253,13 @@ async def _run_whole_call_qa_analysis(
     metrics = compute_call_metrics(rtf_events, call_duration)
 
     # Resolve LLM config
-    system_prompt = qa_node_data.get("qa_system_prompt", "")
+    system_prompt = qa_data.qa_system_prompt or ""
     if not system_prompt:
         logger.warning("No system prompt defined for QA Node")
         return {"error": "no_system_prompt", "node_results": {}}
 
     provider, model, api_key, service_kwargs = await resolve_llm_config(
-        qa_node_data, workflow_run
+        qa_data, workflow_run
     )
 
     if not api_key:

api/services/workflow/qa/llm_config.py

@@ -4,10 +4,11 @@ import random
 
 from api.db import db_client
 from api.db.models import WorkflowRunModel
+from api.services.workflow.dto import QANodeData
 
 
 async def resolve_llm_config(
-    qa_node_data: dict, workflow_run: WorkflowRunModel
+    qa_data: QANodeData, workflow_run: WorkflowRunModel
 ) -> tuple[str, str, str, dict]:
     """Resolve the LLM provider, model, API key, and extra kwargs for QA analysis.
 
@@ -18,24 +19,23 @@ async def resolve_llm_config(
         (provider, model, api_key, service_kwargs) tuple — service_kwargs can be
         passed directly to create_llm_service_from_provider as keyword arguments.
     """
-    if not qa_node_data.get("qa_use_workflow_llm", True):
-        provider = qa_node_data.get("qa_provider", "openai")
+    if not qa_data.qa_use_workflow_llm:
+        provider = qa_data.qa_provider or "openai"
         kwargs = {}
         if provider == "azure":
-            kwargs["endpoint"] = qa_node_data.get("qa_endpoint", "")
+            kwargs["endpoint"] = qa_data.qa_endpoint or ""
         return (
             provider,
-            qa_node_data.get("qa_model"),
-            qa_node_data.get("qa_api_key"),
+            qa_data.qa_model,
+            qa_data.qa_api_key,
             kwargs,
         )
 
     # Fall back to user's configured LLM
     provider, model, api_key, kwargs = await resolve_user_llm_config(workflow_run)
 
-    qa_model = qa_node_data.get("qa_model", "default")
-    if qa_model and qa_model != "default":
-        model = qa_model
+    if qa_data.qa_model and qa_data.qa_model != "default":
+        model = qa_data.qa_model
 
     return provider, model, api_key, kwargs
 

api/services/workflow/qa/node_summary.py

@@ -7,7 +7,7 @@ from loguru import logger
 from api.db import db_client
 from api.db.models import WorkflowRunModel
 from api.services.pipecat.service_factory import create_llm_service_from_provider
-from api.services.workflow.dto import NodeType
+from api.services.workflow.dto import NodeType, QANodeData
 from api.services.workflow.qa.llm_config import resolve_llm_config
 from api.services.workflow.qa.tracing import create_node_summary_trace
 from pipecat.processors.aggregators.llm_context import LLMContext
@@ -48,7 +48,7 @@ async def ensure_node_summaries(
     workflow_definition: dict,
     definition_id: int | None,
     workflow_run: WorkflowRunModel,
-    qa_node_data: dict,
+    qa_data: QANodeData,
 ) -> dict[str, Any]:
     """Ensure every agentNode/startCall node has a summary in the definition.
 
@@ -69,7 +69,7 @@ async def ensure_node_summaries(
         return existing_summaries
 
     provider, model, api_key, service_kwargs = await resolve_llm_config(
-        qa_node_data, workflow_run
+        qa_data, workflow_run
     )
     if not api_key:
         logger.warning("No API key for node summary generation, skipping")