Fed Speech MCP

An MCP (Model Context Protocol) server that retrieves, parses, and analyzes speeches and testimonies from major Federal Reserve officers.

Features

• 📡 RSS Feed Discovery - Automatically discovers new speeches from Fed RSS feeds
• 🔍 Index Page Scanning - Backfill capability by scanning yearly index pages
• 📝 Smart Parsing - Extracts metadata, speaker info, and clean text from Fed HTML pages
• 📊 Feature Extraction - Detects topics (inflation, rates, labor market, etc.)
• ⚖️ Importance Scoring - Rule-based scoring for market relevance
• 💾 JSON Storage - Persistent storage with deduplication
• 🔌 MCP Interface - Full MCP compatibility for AI assistants

Covered Content

Speakers

• Chair of the Federal Reserve
• Vice Chair of the Federal Reserve
• Federal Reserve Governors

Content Types

• Speeches
• Congressional Testimony
• Prepared Remarks

Installation

Prerequisites

• Python 3.10 or higher
• uv package manager

Install uv (if you don't have it):

# macOS / Linux
curl -LsSf https://astral.sh/uv/install.sh | sh

# Windows
powershell -ExecutionPolicy ByPass -c "irm https://astral.sh/uv/install.ps1 | iex"

Install the Project

cd fed-speech-mcp

# Create virtual environment and install
uv sync

That's it! The virtual environment is created and all dependencies are installed.

🚀 Quick Start: Test Locally in 5 Minutes

Want to see the MCP tools in action before connecting to an AI? Follow these steps:

Step 1: Install

cd fed-speech-mcp
uv sync

Step 2: Fetch Speeches from Fed Website

uv run python scripts/test_local.py refresh --limit 5

You'll see output like:

🔄 REFRESHING SPEECHES FROM FED WEBSITE
==================================================
📡 Discovering documents via RSS feeds...
   Found 20 document(s)

[1/5] Processing: Governor Bowman's Speech on Banking...
   ✅ Fetched 45231 bytes
   ✅ Parsed: Michelle W. Bowman - speech
   ✅ Normalized: fed-speech-a1b2c3d4e5f6
   ✅ Saved (NEW)
...
✅ REFRESH COMPLETE
   Discovered: 5
   New: 5
   Total in storage: 5

Step 3: Explore the Data

View latest speeches:

uv run python scripts/test_local.py latest --limit 3

Output:

📰 LATEST SPEECHES
==================================================
Found 3 speech(es):

1. Speech on Monetary Policy and the Economic Outlook
   📅 2024-12-15
   👤 Jerome H. Powell (Chair)
   📄 speech | 2500 words
   ⭐ high (0.85)
   🏷️  Topics: inflation, rates, labor
   🔗 https://federalreserve.gov/...
   🆔 fed-speech-abc123

Search by keyword:

uv run python scripts/test_local.py search --query "inflation"

Filter by speaker:

uv run python scripts/test_local.py speaker --name "Powell"
uv run python scripts/test_local.py speaker --role "Chair"

Get full speech content:

uv run python scripts/test_local.py get --doc-id fed-speech-abc123

View statistics:

uv run python scripts/test_local.py stats

Step 4: Run All Tests

uv run python scripts/test_local.py all

This runs a complete test suite: refresh → stats → latest → search → filter → get.

Available Test Commands

Using with AI Assistants

Platform Compatibility Overview

Step-by-Step: Claude Desktop

Claude Desktop has native MCP support. This is the easiest setup.

Step 1: Install the Package

cd fed-speech-mcp
uv sync

Step 2: Locate Claude Desktop Config

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

Step 3: Add MCP Server Configuration

Edit claude_desktop_config.json:

{
  "mcpServers": {
    "fed-speech": {
      "command": "uv",
      "args": ["run", "fed-speech-mcp"],
      "cwd": "/path/to/fed-speech-mcp"
    }
  }
}

Note: Replace /path/to/fed-speech-mcp with your actual installation path.

Step 4: Restart Claude Desktop

Close and reopen Claude Desktop. You should see the Fed Speech tools available.

Step 5: Start Using

Ask Claude things like:

• "Refresh the Fed speeches and show me the latest ones"
• "What did Chair Powell say about inflation recently?"
• "Show me all testimony from 2024"

Step-by-Step: Cursor IDE

Cursor IDE also supports MCP natively.

Step 1: Install the Package

cd fed-speech-mcp
uv sync

Step 2: Open Cursor Settings

1. Open Cursor IDE
2. Go to Settings (⌘+, on macOS, Ctrl+, on Windows)
3. Search for "MCP" or navigate to Features → MCP Servers

Step 3: Add MCP Server

Click "Add MCP Server" and configure:

{
  "name": "fed-speech",
  "command": "uv",
  "args": ["run", "fed-speech-mcp"],
  "cwd": "/path/to/fed-speech-mcp"
}

Step 4: Enable and Use

Toggle the server on. You can now use Fed Speech tools in Cursor's AI chat.

Step-by-Step: ChatGPT (Custom GPT)

ChatGPT doesn't support MCP natively, but you can use the HTTP API wrapper.

Step 1: Install the Package

cd fed-speech-mcp
uv sync

Step 2: Start the HTTP API Server

uv run fed-speech-http

The server runs at http://localhost:8000 by default.

Step 3: Expose to Internet (Required for ChatGPT)

Use a tunneling service like ngrok:

# Install ngrok: https://ngrok.com/download
ngrok http 8000

Copy the public URL (e.g., https://abc123.ngrok.io).

Step 4: Create a Custom GPT

1. Go to ChatGPT

2. Click your profile → My GPTs → Create a GPT

3. In the Configure tab:

   • Name: "Fed Speech Analyst"
   • Description: "Analyzes Federal Reserve speeches and testimonies"
   • Instructions:

     You are a Federal Reserve speech analyst. Use the available actions to:
     1. Fetch latest speeches with get_latest_speeches
     2. Search speeches by speaker, type, or keywords
     3. Analyze speech content for market-relevant insights
     Always refresh speeches first if the user asks about recent content.
     

4. Click Create new action and import the OpenAPI schema:

openapi: 3.0.0
info:
  title: Fed Speech API
  version: 1.0.0
servers:
  - url: https://your-ngrok-url.ngrok.io
paths:
  /speeches/latest:
    get:
      operationId: getLatestSpeeches
      summary: Get latest Fed speeches
      parameters:
        - name: limit
          in: query
          schema:
            type: integer
            default: 10
        - name: since_date
          in: query
          schema:
            type: string
      responses:
        '200':
          description: List of speeches
  /speeches/search:
    get:
      operationId: searchSpeeches
      summary: Search speeches by keyword
      parameters:
        - name: query
          in: query
          required: true
          schema:
            type: string
        - name: limit
          in: query
          schema:
            type: integer
            default: 10
      responses:
        '200':
          description: Search results
  /speeches/{doc_id}:
    get:
      operationId: getSpeech
      summary: Get a specific speech
      parameters:
        - name: doc_id
          in: path
          required: true
          schema:
            type: string
      responses:
        '200':
          description: Speech details
  /speeches/refresh:
    post:
      operationId: refreshSpeeches
      summary: Fetch new speeches from Fed website
      responses:
        '200':
          description: Refresh result

Step 5: Save and Use

Save your Custom GPT. Now you can ask it questions about Fed speeches!

Step-by-Step: Google Gemini

Google Gemini can use the HTTP API via Google AI Studio or API.

Step 1: Start the HTTP API Server

cd fed-speech-mcp
uv sync
uv run fed-speech-http

Step 2: Expose to Internet

ngrok http 8000

Step 3: Use with Google AI Studio

1. Go to Google AI Studio
2. Create a new prompt or chat
3. In your system instructions, include:

You have access to a Fed Speech API at https://your-ngrok-url.ngrok.io

Available endpoints:
- GET /speeches/latest?limit=10 - Get latest speeches
- GET /speeches/search?query=inflation - Search speeches
- GET /speeches/{doc_id} - Get specific speech
- POST /speeches/refresh - Fetch new speeches
- GET /speeches/by-speaker?name=Powell - Filter by speaker
- GET /speeches/by-type?doc_type=testimony - Filter by type

When users ask about Fed speeches, use these endpoints to fetch data.
Format responses clearly with speaker names, dates, and key points.

Step 4: Use Function Calling (Advanced)

For programmatic use with Gemini API:

import google.generativeai as genai
import requests

# Configure Gemini
genai.configure(api_key="YOUR_API_KEY")

# Define tools for Gemini
tools = [
    {
        "function_declarations": [
            {
                "name": "get_latest_speeches",
                "description": "Get the latest Federal Reserve speeches",
                "parameters": {
                    "type": "object",
                    "properties": {
                        "limit": {"type": "integer", "description": "Max speeches to return"}
                    }
                }
            },
            {
                "name": "search_speeches",
                "description": "Search Fed speeches by keyword",
                "parameters": {
                    "type": "object",
                    "properties": {
                        "query": {"type": "string", "description": "Search query"}
                    },
                    "required": ["query"]
                }
            }
        ]
    }
]

model = genai.GenerativeModel('gemini-pro', tools=tools)

# Handle function calls
def handle_function_call(fn_name, args):
    base_url = "http://localhost:8000"
    if fn_name == "get_latest_speeches":
        resp = requests.get(f"{base_url}/speeches/latest", params=args)
    elif fn_name == "search_speeches":
        resp = requests.get(f"{base_url}/speeches/search", params=args)
    return resp.json()

# Chat with function calling
chat = model.start_chat()
response = chat.send_message("What are the latest Fed speeches about inflation?")

# Process function calls in response
for part in response.parts:
    if hasattr(part, 'function_call'):
        fn = part.function_call
        result = handle_function_call(fn.name, dict(fn.args))
        # Send result back to model
        response = chat.send_message(str(result))

HTTP API Reference

When using the HTTP API wrapper, these endpoints are available:

Query Parameters

• limit - Max results (default: 10, max: 50)
• since_date - ISO 8601 date filter
• start_date / end_date - Date range
• name - Speaker name (partial match)
• role - "Chair", "Vice Chair", or "Governor"
• doc_type - "speech", "testimony", or "prepared_remarks"
• query - Search keywords

Running the MCP Server Directly

For native MCP clients:

# Run the MCP server
uv run fed-speech-mcp

# Run the HTTP API server
uv run fed-speech-http

Available Tools

get_latest_speeches

Get the latest Federal Reserve speeches, sorted by publication date.

Parameters:

• limit (optional): Maximum number of speeches (default: 10, max: 50)
• since_date (optional): Only return speeches after this date (ISO 8601)

get_speeches_by_speaker

Filter speeches by speaker name and/or role.

Parameters:

• name (optional): Speaker name (partial match, e.g., "Powell")
• role (optional): "Chair", "Vice Chair", or "Governor"
• start_date (optional): Start date filter
• end_date (optional): End date filter

get_speeches_by_type

Get speeches by document type.

Parameters:

• doc_type (required): "speech", "testimony", or "prepared_remarks"
• start_date (optional): Start date filter
• end_date (optional): End date filter

get_speech

Get full content and metadata for a specific speech.

Parameters:

• doc_id (required): The unique document identifier

refresh_speeches

Fetch new speeches from the Federal Reserve website.

Parameters:

• include_index (optional): Also scan index pages (slower but thorough)
• years (optional): Years to scan for index pages

search_speeches

Search speeches by keyword.

Parameters:

• query (required): Search query
• limit (optional): Maximum results (default: 10)

get_speech_stats

Get statistics about stored speeches.

Output Format

Each speech document contains:

{
  "doc_id": "fed-speech-abc123def456",
  "source": {
    "publisher": "Board of Governors of the Federal Reserve System",
    "collection": "speeches",
    "url": "https://www.federalreserve.gov/...",
    "retrieved_at": "2024-01-15T10:30:00Z"
  },
  "published_at": "2024-01-15T00:00:00Z",
  "title": "Speech Title",
  "speaker": {
    "name": "Jerome H. Powell",
    "role": "Chair",
    "organization": "Board of Governors of the Federal Reserve System"
  },
  "doc_type": "speech",
  "event": {
    "name": "Economic Club of New York",
    "location": "New York, NY"
  },
  "text": {
    "raw": "...",
    "clean": "..."
  },
  "features": {
    "word_count": 2500,
    "language": "en",
    "has_qa": false,
    "topics": {
      "inflation": true,
      "labor_market": true,
      "rates": true,
      "balance_sheet": false,
      "growth": true,
      "financial_stability": false
    }
  },
  "importance": {
    "tier": "high",
    "score": 0.85,
    "reasons": [
      "Speaker is Chair (Jerome H. Powell)",
      "Discusses rates in context of inflation"
    ]
  }
}

Importance Scoring

The importance score is calculated using these rules:

Topic Detection

Topics are detected by keyword matching:

• Inflation: inflation, prices, CPI, PCE, price stability
• Labor Market: employment, unemployment, wages, jobs
• Rates: interest rate, fed funds, hike, cut, monetary policy
• Balance Sheet: QE, QT, runoff, asset purchases
• Growth: GDP, demand, recession, economic activity
• Financial Stability: banking, liquidity, stress, systemic risk

Environment Variables

Data Storage

data/
├── speeches/           # Processed JSON documents
│   ├── fed-speech-xxx.json
│   └── ...
└── raw/               # Raw HTML content (for traceability)
    ├── 20240115_abc123.html
    └── ...

Development

Running Unit Tests

# Install with dev dependencies
uv sync --dev

# Run all tests
uv run pytest

# Run with verbose output
uv run pytest -v

# Run specific test file
uv run pytest tests/test_models.py

# Run with coverage
uv run pytest --cov=fed_speech_mcp

Running Local Integration Tests

# Test all MCP functionality end-to-end
uv run python scripts/test_local.py all

# Test specific functionality
uv run python scripts/test_local.py refresh --limit 3
uv run python scripts/test_local.py search --query "rates"

Project Structure

fed-speech-mcp/
├── src/fed_speech_mcp/
│   ├── __init__.py
│   ├── server.py          # MCP server entry point
│   ├── http_server.py     # HTTP API wrapper
│   ├── config.py          # Configuration
│   ├── models/            # Pydantic data models
│   ├── ingestion/         # RSS/index discovery & fetching
│   ├── parsing/           # HTML parsing & normalization
│   ├── features/          # Feature extraction & scoring
│   └── storage/           # JSON storage layer
├── data/                  # Data directory
├── pyproject.toml
└── README.md

License

MIT

Acknowledgments

Data sourced from the Federal Reserve Board.