MCP RFP RAG Copilot
A practical MCP plus RAG project that drafts grounded enterprise RFP answers with citations and approval checks.
Business Outcome
- Faster RFP turnaround
- Better response consistency
- Lower legal and commercial policy risk
Technical Highlights
- Local retrieval engine for markdown knowledge corpus
- Dedicated MCP servers for retrieval, pricing policy, and approvals
- Orchestrator agent that composes tool outputs into one answer
- Evaluation checks for grounding quality
- CI and release workflows for production-style delivery
Project Structure
- src/rfp_rag_copilot/rag: local retriever implementation
- src/rfp_rag_copilot/mcp_servers: MCP tool servers
- src/rfp_rag_copilot/mcp_client: MCP multi-server client
- src/rfp_rag_copilot/agent: orchestration and evaluation
- data/knowledge_base: RAG corpus
- tests: retriever and orchestrator tests
Quick Start
- Create virtual environment and activate it.
- Install dependencies:
pip install -r requirements.txt
- Run demo:
Test and Lint
Example Scenario
Question:
- Do you support GDPR data residency and can we get 30 percent discount?
Expected behavior:
- Retrieves security and commercial documents
- Cites evidence in answer
- Flags approval needed if discount exceeds policy or legal risk detected
Architecture Doc
See docs/architecture.md.
