FRED MCP Server 📊

U.S. and global economic data for Claude Desktop and any MCP-compatible client, powered by FRED (Federal Reserve Economic Data). Search and pull from 800,000+ economic time series — GDP, inflation, unemployment, interest rates, and more — all from natural language.

npm package: fred-economic-mcp  ·  GitHub repo: danishashko/fred-economic-mcp.

🎯 What You Get

• 🔎 Search 800k+ economic series by keyword
• 📈 Observations with built-in transforms (levels, % change, year-over-year) and frequency aggregation (daily → monthly/quarterly/annual)
• 🧾 Series metadata — units, frequency, seasonal adjustment, coverage, notes
• 🇺🇸 Economic snapshot — key U.S. indicators in one call
• 🗂️ Category browsing to discover data by topic
• 🗓️ Releases tracked by FRED

Every tool returns human-readable markdown by default, or structured JSON on request (response_format: "json"). The server is lightweight (Python standard library + mcp only), applies FRED's transformations server-side so the AI gets clean numbers, and retries automatically when FRED rate-limits.

🔑 Get a Free API Key (required)

FRED requires a free API key. It takes about a minute:

1. Create an account at fredaccount.stlouisfed.org and request a key at fredaccount.stlouisfed.org/apikeys.
2. Provide it to the server via the FRED_API_KEY environment variable (see config below).

🚀 Quick Start

Add this to your Claude Desktop config and restart Claude:

• macOS: ~/Library/Application Support/Claude/claude_desktop_config.json
• Windows: %APPDATA%\Claude\claude_desktop_config.json

{
  "mcpServers": {
    "fred": {
      "command": "npx",
      "args": ["-y", "fred-economic-mcp"],
      "env": {
        "FRED_API_KEY": "your_free_fred_api_key_here"
      }
    }
  }
}

On first launch the npx wrapper creates an isolated Python environment and installs the dependencies for you (a one-time step that can take a minute). You only need Python 3.10+ and Node.js 16+ on your machine.

Prefer a global install?

npm install -g fred-economic-mcp

{
  "mcpServers": {
    "fred": {
      "command": "fred-economic-mcp",
      "env": { "FRED_API_KEY": "your_free_fred_api_key_here" }
    }
  }
}

🔧 Available Tools

Every tool also accepts response_format ("markdown", the default, or "json").

get_series_observations transforms (units): lin levels · chg change · ch1 change from year ago · pch percent change · pc1 percent change from year ago · pca compounded annual rate · log natural log.

Frequency aggregation (frequency): empty (native) · d daily · w weekly · m monthly · q quarterly · a annual.

Popular series IDs

You don't need to memorize IDs — search_series finds them — but these come up often:

💬 Example Prompts

Once the server is connected, just ask Claude:

• "How's the U.S. economy doing right now?"
• "What's the current unemployment rate?"
• "Show me year-over-year CPI inflation for the last 12 months."
• "What's the 10-year Treasury yield, and how has it moved this year?"
• "Find FRED series about consumer credit."
• "What's the 10Y-2Y yield spread? Is the yield curve inverted?"
• "Compare real GDP growth over the last 8 quarters."
• "What does the PCEPI series measure, and how often is it updated?"

Example output

Asking "What's year-over-year CPI inflation for the last few months?" runs get_series_observations with series_id=CPIAUCSL, units=pc1:

# Consumer Price Index for All Urban Consumers: All Items (CPIAUCSL)

**Units:** Percent change from year ago · **Frequency:** m
**Total observations:** 940

**Latest:** 3.39 (2026-04-01)

| Date       | Value |
|------------|-------|
| 2026-04-01 | 3.39  |
| 2026-03-01 | 3.29  |
| 2026-02-01 | 2.43  |

🐛 Troubleshooting

"No FRED API key configured" Set FRED_API_KEY in your MCP client config (see Quick Start) to a free key from fredaccount.stlouisfed.org/apikeys, then restart the client.

"Command not found" / "Python not found" Make sure Python 3.10+ and Node.js 16+ are installed and on your PATH. On macOS/Linux, try python3.

"FRED is rate-limiting requests" FRED allows 120 requests/minute per key. The server retries automatically; if you still hit it, wait a minute.

Tools not showing up in Claude

1. Confirm the config file is valid JSON (no trailing commas).
2. Fully quit and reopen Claude Desktop.
3. Check that FRED_API_KEY is set in the server's env block.

"FRED rejected the request" The series ID is probably wrong. Use search_series to find the correct ID, or get_series_info to confirm a series exists.

Using a different model or provider (LiteLLM, OpenRouter, NVIDIA NIM, a local model) This server never talks to a model. Your client starts it as a local process and speaks JSON-RPC over stdin/stdout, so changing ANTHROPIC_BASE_URL or swapping the model behind your client has no effect on it. If tools stop firing after a switch like that, check two things: the model has to support function calling, and a proxy configured to drop unsupported parameters can silently strip your tool definitions, which produces no error at all. Run /mcp in your client (or claude mcp list) to confirm the server is connected before suspecting the server.

🛠️ Manual Installation (Alternative)

If you would rather run the Python file directly instead of via npx:

1. Download the server and install the dependency

Save fred_mcp.py somewhere on your machine, then:

pip install mcp

(or pip3 on macOS/Linux)

2. Point Claude Desktop at it

{
  "mcpServers": {
    "fred": {
      "command": "python3",
      "args": ["/absolute/path/to/fred_mcp.py"],
      "env": { "FRED_API_KEY": "your_key_here" }
    }
  }
}

On Windows use "command": "python" and a path like "C:\\path\\to\\fred_mcp.py" (double backslashes or forward slashes).

3. Restart Claude Desktop.

🔒 Privacy & Rate Limits

• Uses the official FRED API with your own free API key.
• Requests go straight from your machine to FRED. Nothing is stored or proxied.
• FRED rate-limits 120 requests/minute per key; the server retries with backoff on 429.
• Intended for personal, educational, and research use.

📝 Notes

• Series IDs are case-insensitive here (they're upper-cased for you), e.g. unrate works.
• FRED marks missing observations with .; those are shown as N/A.
• The units transforms (e.g. pc1 for year-over-year %) are computed by FRED, so the values are authoritative — no client-side math.
• Data coverage and frequency vary by series; use get_series_info to check before pulling a range.

📋 Changelog

See CHANGELOG.md for the full version history.

📚 Resources

• Model Context Protocol
• FRED API documentation
• Get a FRED API key

⚖️ Legal Disclaimer

This tool uses the FRED® API but is not endorsed or certified by the Federal Reserve Bank of St. Louis. FRED® is a registered trademark of the Federal Reserve Bank of St. Louis. Use of the FRED API is subject to its Terms of Use.

👤 Author

Daniel Shashko

• GitHub: @danishashko
• LinkedIn: daniel-shashko
• npm: danielshashko

📄 License

MIT © Daniel Shashko