FMP MCP Research Server

Remote MCP server for ChatGPT that exposes read-only FMP tools to build a strict buy-side earnings-call evidence pack before scoring a catalyst report.

Why this design

The report process requires the agent to read the actual full earnings-call transcript including Q&A, separately review earnings releases and financial tables, document missing evidence, complete a quarter-by-quarter audit, and only then score. This MCP server therefore exposes small, auditable tools instead of a single black-box "make report" tool.

Related MCP server: FMP MCP Server

Tools exposed

FMP endpoints used

Base URL: https://financialmodelingprep.com/stable

• /profile?symbol=...

• /earning-call-transcript-dates?symbol=...

• /earning-call-transcript?symbol=...&year=...&quarter=...

• /earning-call-transcript-latest

• /earnings-calendar?symbol=...&from=...&to=...

• /income-statement?symbol=...&period=quarter&limit=...

• /balance-sheet-statement?symbol=...&period=quarter&limit=...

• /cash-flow-statement?symbol=...&period=quarter&limit=...

• /key-metrics?symbol=...&period=quarter&limit=...

• /ratios?symbol=...&period=quarter&limit=...

• /financial-growth?symbol=...&period=quarter&limit=...

• /sec-filings-search/symbol?symbol=...&from=...&to=...

• /sec-filings-8k?from=...&to=...

Local development

cp .env.example .env
# edit .env and set FMP_API_KEY
python -m venv .venv
source .venv/bin/activate
pip install -e .
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

Deployment recommendation

For ChatGPT, deploy as a remote HTTPS MCP server. Local is best only for development. For production, use Cloud Run or Fly.io for a simple container deployment; use AWS ECS/Fargate if your firm already standardizes on AWS and wants VPC/private networking.

Recommended default: Google Cloud Run

• Container-native and low ops burden.

• Scales to zero for cost control.

• Easy environment variables and secret management.

• Public HTTPS endpoint works with ChatGPT remote MCP setup.

Cloud Run deploy sketch

gcloud run deploy fmp-mcp-research \
  --source . \
  --region us-central1 \
  --allow-unauthenticated \
  --set-env-vars FMP_API_KEY=YOUR_KEY

For a production workspace, replace --allow-unauthenticated with an OAuth/API-gateway layer or Cloud Run IAM where supported by your MCP client path.

ChatGPT connection

Use the public HTTPS URL plus /mcp, for example:

https://your-service-url.run.app/mcp

In ChatGPT app/developer settings, add the custom MCP app/server and approve the read-only tools.

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 full API keys or secrets.

• Add rate limiting if used by multiple analysts.

Typical report workflow

1. research_report_contract(sector="pharma")

2. fmp_build_research_evidence_pack(symbol="AONC", min_year=2025, requested_calls=2)

3. Read each transcript payload, including Q&A.

4. Review statement tables in the evidence pack.

5. Use fmp_search_sec_filings to locate 8-K/10-Q/10-K fallback sources and official release exhibits.

6. Complete the source audit and quarter-by-quarter audit.

7. Only after the audit, create scorecard and adjusted score.