Fed Speech

You can deploy and use the Fed Speech MCP Server to automatically fetch, parse, and analyze speeches and testimonies from Federal Reserve officers. It provides RSS discovery, smart parsing, topic extraction, and an MCP-compatible API so you can query, filter, and analyze content for market-relevant insights.

How to use

You run the MCP server and connect your AI assistant to it. Start the MCP process for parsing and indexing Fed speeches, then use the HTTP API or the local MCP endpoints to refresh data, search by speaker or topic, fetch full speeches, and view statistics. Your AI assistant can request latest speeches, search by keywords, or retrieve a specific document by ID, and you can use the built-in topic and importance scoring to surface material that matters for market context.

How to install

Prerequisites include Python 3.10 or higher and the uv package manager. Install uv if you do not 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"

Additional setup and configuration

Install the Fed Speech MCP project dependencies, then run both the MCP server and the HTTP API server. Use the local HTTP port 8000 by default, or override with FED_SPEECH_HTTP_PORT if you need a different port.

cd fed-speech-mcp
uv sync

Run the MCP server and HTTP API

Start the MCP server and the HTTP API server as separate processes. The MCP server provides the core data ingestion and processing, while the HTTP API exposes endpoints for querying speeches.

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

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

Usage patterns with your AI assistant

• Refresh speeches to fetch new items, backfill from index pages if needed, and deduplicate data.

Notes on environment and endpoints

The HTTP API server runs by default at http://localhost:8000. You can query latest speeches, search by keyword, filter by speaker or type, fetch a specific speech by document ID, and refresh the data store.

Security considerations

Limit access to the HTTP API and MCP endpoints in production. If you expose the API, use authentication or a gateway to restrict usage and monitor requests.

Troubleshooting basics

If you encounter connection issues, verify that both servers are running and listening on the expected ports. Check that the data store directory is writable and that the environment variables listed are correctly set.

Example workflow with an AI assistant

1. Start the MCP and HTTP API servers. 2) Use your AI assistant to request the latest speeches or search for inflation-related remarks. 3) Retrieve a full speech by ID to review details and context. 4) Use the topic and importance scores to assess relevance for market analysis.

Project structure overview

The server includes ingestion for RSS/index discovery, parsing, feature extraction, storage, and an MCP-compatible API surface. You will interact with endpoints to list, search, and fetch speech documents.

Available tools

get_latest_speeches

Retrieve the latest Federal Reserve speeches, with optional limit and since_date filters.

get_speeches_by_speaker

Filter speeches by speaker name and/or role, with optional date filters.

get_speeches_by_type

Get speeches by document type such as speech, testimony, or prepared_remarks.

get_speech

Fetch full content and metadata for a specific speech by document ID.

refresh_speeches

Fetch new speeches from the Federal Reserve website, with optional index scanning.

search_speeches

Search speeches by keyword with optional limit.

get_speech_stats

Retrieve statistics about stored speeches.