The MCP Server Setup

This is the MCP server setup that is used by the agents in ../agentServer. It supports the agents' access to various aspects of the Congress API and the retrieval of relevant sections of a bill (see Section 5.2). We have the following files and directories:

• main.py: Contains the main MCP server functions' logic, and running this file starts the entire MCP server

• main_test.py: A copy of main.py but instead of starting the server, it runs tests that the functions inside main.py are working correctly.

• util/: Contains all the helper functions we use to keep more complex logic away from the main file

• rag/: Contains the entire Retrieval-Augmented Generation (RAG) setup that is used to retrieve the relevant sections of a bill (for now)

• data/: Contains additional data that is used to power the MCP functions

Filling in the secrets.ini file

For the Congress MCP server to work completely, we need API keys for the following services:

• OpenAI (for textual embeddings and LLM generation)
• Congress API (for all the legislative histories of the bills)
• Govinfo.gov (for structured access to the Congressional Record)
• (Optional) Langchain API (to remotely log the retrieval runs)

You can get them here:

• Congress API
• GovInfo API
• OpenAI API
• Langchain API

Note: The Congress API as well as the GovInfo API are free of charge, the OpenAI API is billed per usage.