Finance Data MCP

A Python-first toolkit for SEC filing ingestion, OCR-to-Markdown conversion, transcript collection, and retrieval across hybrid retrieval (dense + BM25) with reranking.

What this project does

• Downloads SEC filings and stores filing metadata.
• Converts filing PDFs to Markdown via olmOCR.
• Chunks and indexes filings/transcripts in Chroma.
• Supports:
  • Hybrid search (dense + BM25 reciprocal-rank-fusion + reranker).
• Exposes workflows through:
  • FastAPI (`server.py`).
  • MCP server (`mcp_server.py`).

Repository layout

• `finance_data/filings/`: SEC download + helpers.
• `finance_data/ocr/`: olmOCR pipeline.
• `finance_data/dataloader/`: chunking, Chroma indexing, semantic + BM25 retrieval.
• `finance_data/earnings_transcripts/`: transcript fetch + persistence.
• `finance_data/server_api/`: API request/response models + batch helpers.
• `server.py`: FastAPI app.
• `mcp_server.py`: MCP entrypoint.
• `docs/`: setup and operations docs.

Quick start

1) Install dependencies

uv sync

For OCR/embedding flows:

uv sync --group ocr-md

For MCP workflows:

uv sync --group ocr-md --group mcp

2) Configure environment

Use `.env` or environment variables. Common settings:

• `SEC_API_ORGANIZATION`, `SEC_API_EMAIL`
• `OLMOCR_SERVER`, `OLMOCR_MODEL`, `OLMOCR_WORKSPACE`
• `EMBEDDING_SERVER`, `EMBEDDING_MODEL`
• `CHROMA_PERSIST_DIR`
• `MCP_HOST`, `MCP_PORT`, `MCP_NGROK_ALLOWED_HOSTS`

See `finance_data/settings.py` for defaults.

3) Run services

Start model servers:

make vllm-olmocr-serve
make vllm-embd-serve
make vllm-reranker-serve

Start API:

make start-server

Start MCP:

uv run --group ocr-md --group mcp python mcp_server.py

Search capabilities

SEC filings API

• Hybrid (dense + BM25 + reranker): `POST /vector_store/search_sec_filings`

Transcript API

• Hybrid (dense + BM25 + reranker): `POST /vector_store/search_transcripts`

MCP tools

• Hybrid: `search_sec_filings_tool`, `search_transcripts_tool`

Core workflows

SEC filing → Markdown

uv run python -m finance_data.filings.sec_data --ticker AMZN --year 2025
uv run python -m finance_data.ocr.olmocr_pipeline --pdf-dir sec_data/AMZN-2025

Embed and search filings (API)

curl -s -X POST "http://127.0.0.1:8081/vector_store/embed_sec_filings" \
  -H "Content-Type: application/json" \
  -d '{"ticker":"AMZN","year":"2025","filing_type":"10-K","force":false}'

curl -s -X POST "http://127.0.0.1:8081/vector_store/search_sec_filings" \
  -H "Content-Type: application/json" \
  -d '{"ticker":"AMZN","year":"2025","filing_type":"10-K","query":"operating income margin","top_k":5}'

Earnings transcripts

Fetch quarterly transcripts:

uv run python -m finance_data.earnings_transcripts.transcripts AMZN 2025

Embed + hybrid search transcripts:

curl -s -X POST "http://127.0.0.1:8081/vector_store/embed_transcripts" \
  -H "Content-Type: application/json" \
  -d '{"ticker":"AMZN","year":"2025","force":false}'

curl -s -X POST "http://127.0.0.1:8081/vector_store/search_transcripts" \
  -H "Content-Type: application/json" \
  -d '{"ticker":"AMZN","year":"2025","query":"AWS revenue growth","top_k":5}'

Docker

Use Makefile wrappers:

make docker-build
make docker-start

Stop/remove by API port:

make docker-stop
make docker-remove

Documentation

• `docs/README.md`
• `docs/setup-and-operations.md`