← All work
PRODUCTIONMCPRAGGovCon

ContractsHub

Federal contract intelligence, live inside Claude and ChatGPT

Finds federal opportunities the moment you ask, profiles vendors and competitors, tracks bid protests, and drafts Pink Team proposal content and capture briefs backed by 28.6M contract awards. 90+ research tools inside ChatGPT and Claude.

28.6M

Contract awards indexed

90+

Research tools exposed

869K

Vendor entities

1.15M

Document chunks embedded

See the real product

Product demo — ContractsHub MCP searching SAM.gov live inside an AI assistant
Product demo — SEC filings, knowledge graph, and GAO protest intelligence tools

The problem

Government contractors burn weeks hand-searching SAM.gov, USASpending, FPDS-NG, and the FAR to find opportunities, vet competitors, and answer compliance questions. AI assistants can't help — they have no live access to any of this data.

What we built

01

Built a FastMCP server that federates five federal data APIs plus Congress.gov behind a single Model Context Protocol endpoint, so Claude, ChatGPT, and Cursor can search opportunities, pull award history, and profile vendors mid-conversation.

02

Engineered two retrieval pipelines: a FAR/DFARS regulatory RAG (~3,000 sections, hybrid BM25 + pgvector semantic search with 1024-dim bge-large embeddings) and a document RAG over solicitation attachments with cross-encoder reranking.

03

Designed the data layer as a separate shared platform: 47 ingestion scripts, a 15-step daily refresh orchestrator with data-quality assertions and anomaly alerts, feeding a 30+ table PostgreSQL schema.

04

Deployed on Azure Container Apps behind an API Management gateway with secret-injection auth, scaling 0→3 replicas — the service costs nearly nothing when idle.

Architecture

Protocol layerFastMCP server, stdio + Streamable HTTP transports, 90+ typed tools
GatewayAzure API Management with backend-secret injection and tiered rate limits
RetrievalHybrid BM25 + pgvector (bge-large-en-v1.5, 1024-dim), cross-encoder reranking
Data platformPostgreSQL + pgvector, 47 ingestion pipelines, daily refresh orchestrator
SourcesSAM.gov, USASpending, FPDS-NG, Federal Register, Congress.gov, GAO protests

Outcomes

  • Live in production on Azure, serving federal opportunity discovery inside Claude with zero copy-paste workflow
  • 28.6M contract awards (FY2021–present), 861K subawards, and 1,799 GAO protest decisions queryable in seconds
  • FAR compliance questions answered with paragraph-level regulatory citations — auditable, not hallucinated
  • Autonomous daily data refresh with per-metric quality assertions; recovers from upstream API failures without intervention

Stack

PythonFastMCPFastAPIPostgreSQLpgvectorsentence-transformersAzure Container AppsAzure APIMDockerRedis