FMP MCP Research Server

Remote MCP server for ChatGPT or another LLM that exposes read-only Financial Modeling Prep tools for professional buy-side earnings-call research workflows.

The server does not generate final investment recommendations. It builds evidence manifests, returns transcript completeness metadata, finds filings/tables, and helps an analyst or LLM decide whether a scorecard is supported by the evidence.

Version 0.3.0 changes

• Aligned the transcript tool contract with the documented workflow: section and max_chars are now supported.
• Fixed truncation-marker detection so normal transcripts are not falsely marked as truncated.
• Added a softer transcript_quality_status: complete, usable_with_warnings, or incomplete.
• Kept Q&A split uncertainty as a warning rather than an automatic hard block.
• Added contract_version and evidence_pack_version.
• Added dev dependencies for pytest, ruff, mypy and respx.
• Added CI skeleton for test/lint.
• Reformatted the project into maintainable Python modules.
• Hardened Docker by running as a non-root user.
• Added explicit statement-table review actions for the latest completed fiscal year and selected quarters.
• Reduced default SEC filing noise in evidence packs by omitting non-core filings unless specifically relevant.

Tools exposed

Tool	Purpose
fmp_get_company_profile	Get sector, industry, market cap and descriptive metadata.
fmp_list_transcript_dates	Discover available earnings-call transcript periods.
fmp_get_earnings_call_transcript	Fetch full transcript, prepared remarks, Q&A, or metadata.
fmp_get_statement_tables	Fetch income statement, balance sheet, cash flow, key metrics, ratios and growth.
fmp_search_sec_filings	Find and prioritize 8-K/6-K earnings releases plus 10-Q/10-K evidence.
fmp_get_earnings_calendar	Confirm earnings dates and EPS actual/estimate context.
fmp_build_research_evidence_pack	Build selected periods, evidence manifest, audit template and next actions.
fmp_validate_research_evidence	Mechanically validate an evidence-pack payload.
research_report_contract	Return report sections, audit fields and score dimensions.

Recommended workflow

1. Call research_report_contract(sector="healthcare_technology") or the relevant sector.
2. Call fmp_build_research_evidence_pack(symbol="PSNL", min_year=2025, requested_calls=2).
3. Read selected_periods, evidence_manifest, scoring_readiness and recommended_next_actions.
4. For every selected period, call fmp_get_earnings_call_transcript(symbol, year, quarter, section="full").
5. If content_truncated_by_tool=true, fetch section="prepared_remarks" and section="qna" separately.
6. Mark full_call_text_read=yes and qna_reviewed=yes only after actually reading the returned text.
7. Use fmp_search_sec_filings to locate official 8-K/6-K releases and 10-Q/10-K filings.
8. Use fmp_get_statement_tables(period="annual") to review Income Statement, Balance Sheet and Cash Flow Statement for the latest completed fiscal year.
9. Use fmp_get_statement_tables(period="quarter") to review Income Statement, Balance Sheet and Cash Flow Statement for every selected quarter.
10. Use key metrics, ratios and growth tables as supporting context, not as a substitute for primary statements.
11. Complete both source_audit_template and financial_statement_audit_template before producing a scorecard.

Local development

cp .env.example .env
# edit .env and set FMP_API_KEY
python -m venv .venv
source .venv/bin/activate
pip install -e ".[dev]"
pytest
python -m fmp_mcp_research.server

The MCP endpoint is usually available at:

http://localhost:8000/mcp

Docker

cp .env.example .env
# edit .env and set FMP_API_KEY
docker compose up --build

Security guardrails

• Keep FMP_API_KEY server-side only.
• Do not expose write tools.
• Do not add a tool that generates final investment recommendations.
• Log tool name, symbol, quarter and source coverage status; do not log API keys.
• Add rate limiting and an auth layer if used by multiple analysts or exposed outside a trusted network.