mcp-agentic-pipelines

One MCP server. Five AI pipelines. Thirty-one tools. Zero manual setup.

Composable, agentic AI pipelines for Anthropic's Model Context Protocol. Fact-check claims with DSPy, search documents with hybrid RAG, extract intelligence with DeepPipe, and run clinical voice intake — all from a single MCP server. Designed for Claude Desktop, VS Code, Cursor, and any MCP-compatible client.

---

All 31 Tools

🔍 DeepPipe — Document Intelligence (10 tools)

Tool	What it does
deeppipe_search	Full-text + vector hybrid search across all documents
deeppipe_ingest	Ingest raw text/JSON documents for indexing
deeppipe_ingest_file	Ingest documents directly from file paths
deeppipe_chat_context	RAG-style Q&A with source citations
deeppipe_extractive_answer	Extract a precise answer from documents
deeppipe_list_documents	List all indexed documents with metadata
deeppipe_get_document	Retrieve a specific document by ID
deeppipe_get_text	Get the full text content of a document
deeppipe_remove_document	Remove a document from the index
deeppipe_stats	View index statistics and document count

🧠 Piste — Fact-Checking (6 tools)

Tool	What it does
piste_fact_check	Run a claim through the full 4-stage DSPy pipeline (retrieval → verification → aggregation → verdict)
piste_list_verdicts	Browse all past fact-check verdicts
piste_replay	Replay the audit trail of any fact-check run
piste_get_audit	Get the detailed reasoning chain for a verdict
piste_get_verdict	Retrieve a specific verdict by claim ID
piste_submit_feedback	Submit human feedback to improve future checks

📚 Precis — RAG Pipeline (8 tools)

Tool	What it does
precis_query	Full RAG query — hybrid search + LLM answer generation
precis_list_documents	List all documents in the RAG corpus
precis_debug_stem	Inspect how the stemmer processes a query
precis_debug_search	See raw hybrid search results (before LLM)
precis_upload_document	Upload a document to the RAG corpus
precis_upload_batch	Batch-upload multiple documents at once
precis_extract_work_order	Extract structured work orders from documents
precis_list_work_orders	Browse all extracted work orders

🎙️ Clinical — Voice Intake (5 tools)

Tool	What it does
clinical_start_session	Start a new voice intake session (returns session ID)
clinical_process_audio	Send audio → STT transcription → SOAP notes
clinical_generate_podcast	Generate a clinical podcast from session notes
clinical_list_sessions	List all voice intake sessions
clinical_get_session	Get full details of a specific session

⚙️ Built-in (2 tools)

Tool	What it does
mcp_health	Server health, tool count, provider status
mcp_list_providers	List all 9 supported LLM providers

---

Compose Pipelines Together

These tools are designed to be chained. Here's a real workflow:

1. deeppipe_ingest           → Load a research paper into the index
2. precis_upload_document    → Add it to the RAG corpus
3. deeppipe_search           → Find relevant passages
4. piste_fact_check          → Verify claims found in those passages
5. precis_query              → Generate a RAG answer from verified context
6. clinical_start_session    → Start voice notes on the findings
7. clinical_generate_podcast → Turn it all into a shareable podcast

In Claude Desktop, this is a conversation:

"Search my documents for claims about climate policy, fact-check each one, then generate a clinical podcast summarizing the verified findings."

Claude or

... View full README on GitHub