self-mcp-server

FastMCP server for querying FRED financial and macroeconomic data.

Design doc:

• FRED FastMCP Server 设计文档
• MCP Server 部署目录说明

Setup

uv sync
$env:FRED_API_KEY = ""
uv run python main.py

Alternatively, create a local .env file from .env.example:

Copy-Item .env.example .env

Then set FRED_API_KEY in .env. The server loads it from environment variables via python-dotenv; the key is never exposed as a tool parameter.

Local MCP Loader Demo

This repo also includes a small local MCP loader demo. The config is in deployments/mcp.local.yaml and follows the same simple stdio shape used by Codex/OpenCode-style MCP configs: server name plus command, args, env, and optional env_files.

The demo loads:

• fred: this local FRED MCP server via .venv/Scripts/python.exe main.py
• github: the local Docker image docker.xuanyuan.run/mcp/github:latest

Prepare local secrets:

Copy-Item .env.example .env
Copy-Item deployments/github-mcp/.env.example deployments/github-mcp/.env

Set FRED_API_KEY in .env and GITHUB_PERSONAL_ACCESS_TOKEN in deployments/github-mcp/.env.

Run the demo:

.venv\Scripts\python.exe scripts\demo_load_mcp_servers.py

It initializes both MCP servers, lists their tools, and calls one sample tool on each server.

For HTTP mode, these can also live in .env:

MCP_TRANSPORT=http
MCP_HOST=127.0.0.1
MCP_PORT=8765
MCP_PATH=/mcp
MCP_STATELESS_HTTP=true
MCP_JSON_RESPONSE=true

HTTP mode for curl-based E2E checks:

$env:MCP_TRANSPORT = "http"
$env:MCP_HOST = "127.0.0.1"
$env:MCP_PORT = "8765"
$env:MCP_PATH = "/mcp"
uv run python main.py

Tools

• fred_search_series
• fred_get_series
• fred_get_observations
• fred_get_latest_observation
• fred_get_vintage_dates
• fred_get_canonical_indicators
• fred_get_macro_snapshot

All tools return a common envelope:

{
  "ok": true,
  "data": {},
  "meta": {
    "source": "FRED",
    "endpoint": "series/observations",
    "cached": false,
    "fetched_at": "2026-05-13T12:00:00Z"
  },
  "request": {},
  "warnings": []
}