US Gov Open Data MCP

US Government Open Data MCP

I was listening to things like the State of the Union and hearing numbers thrown around from news articles, from the left, from the right, from everyone. I kept wanting to actually verify what was being said or at least get more context around it. The problem was that the data is spread across dozens of different government agencies with different APIs, different authentication methods, and different formats.

If you've ever tried to look something up on a government site, you probably know the experience. The information is there but actually getting to it is a whole other story. I found myself constantly jumping between sites trying to piece things together and thought

There has to be a better way to do this.

So, I built an MCP server that connects to 37 different U.S. government and international data APIs. It currently has 198 tools covering things like economic data, health statistics, campaign finance, lobbying records, patents, energy, education, and a lot more. The whole idea is that this information should be transparent and easily accessible for people.

The whole thing is built with FastMCP, has disk caching so you're not hammering the APIs with duplicate requests, rate limiting to stay within the various API limits, and automatic retries with backoff for when things fail.

What is MCP?

For those who don't know, MCP (Model Context Protocol) is basically a standard that lets AI assistants call external tools. So instead of the AI just guessing at an answer about the national debt or unemployment rate, it can actually go query the real data from the source and give you an accurate response with actual numbers.

Why I built it

As I mentioned, this information is public and paid for by taxpayers. I figured if I could make it easier for myself to look things up and cross reference what I was hearing then maybe it could help others do the same.

Manually going to each agency's site was tedious and the APIs all have their own quirks. Some use REST, some use GraphQL-style queries, some have weird pagination, and nearly all of them have different error handling. Wrapping all of them in a consistent interface seemed like the right move.

What started as "let me just wrap the Treasury API real quick" turned into 37 APIs and 198 tools. I kept finding more sources and thinking "oh that would be useful too" and here we are. AI was a huge help throughout the process for dealing with the sheer number of different API formats and documentation styles. As with my other projects I see it as a tool, not a replacement, and there was plenty of manual work figuring out how each API actually behaves versus what their docs say it does.

Example Analyses

To show what this thing can actually do I put together a few example analyses. These were generated entirely from live government API data with no manual data entry.

Data Sources

Here is everything it currently covers. Some of these have multiple tools under them so the actual number of things you can query is quite a bit higher than this list makes it seem.

19 of these APIs require no key at all. The rest need free keys that most take under a minute to get.

How to use it

The quickest way to get started is just running it with npx.

npx us-gov-open-data-mcp

If you're using VS Code with Copilot you can add it to your settings like this:

{
  "servers": {
    "us-gov-open-data": {
      "command": "npx",
      "args": ["-y", "us-gov-open-data-mcp"],
      "env": {
        "FRED_API_KEY": "your_key",
        "DATA_GOV_API_KEY": "your_key"
        // ... other api keys
      }
    }
  }
}

For Claude Desktop you can add it to your claude_desktop_config.json:

{
  "mcpServers": {
    "us-gov-open-data": {
      "command": "npx",
      "args": ["-y", "us-gov-open-data-mcp"],
      "env": {
        "FRED_API_KEY": "your_key"
      }
    }
  }
}

It also supports HTTP streaming if you want to use it with web apps or remote access:

node dist/server.js --transport httpStream --port 8080

Selective Module Loading

You don't have to load all 37 APIs if you don't need them. You can pick and choose which modules you want for faster startup and a smaller tool list.

# Via CLI flag
node dist/server.js --modules fred,treasury,congress

# Via environment variable
MODULES=fred,bls,treasury node dist/server.js

The SDK

One thing I wanted to make sure of is that you don't have to use MCP to get value out of this. I built a standalone TypeScript SDK that you can import directly into your own projects.

// Individual module imports
import { getObservations } from "us-gov-open-data-mcp/sdk/fred";
import { getIndicator } from "us-gov-open-data-mcp/sdk/world-bank";
import { getLeadingCausesOfDeath } from "us-gov-open-data-mcp/sdk/cdc";
import { searchBills } from "us-gov-open-data-mcp/sdk/congress";

// Or barrel import everything
import * as sdk from "us-gov-open-data-mcp/sdk";

// FRED
const gdp = await getObservations("GDP", { sort: "desc", limit: 5 });
console.log(gdp.observations);

// World Bank - compare U.S. to Germany
const health = await getIndicator("SH.XPD.CHEX.PC.CD", {
  country: "US;DE",
  dateRange: "2015:2023"
});

// CDC
const deaths = await getLeadingCausesOfDeath({
  state: "California",
  year: 2017
});

All SDK functions include disk-backed caching, retry with exponential backoff, and rate limiting. This way if you're building your own app or tool and just want easy access to these APIs you can use the SDK without setting up a server or anything.

API Keys

Here is a quick reference for where to get each key. Again, 19 of the APIs don't need a key at all.

Links

• Docs: https://lzinga.github.io/us-gov-open-data-mcp/
• NPM: us-gov-open-data-mcp
• GitHub: https://github.com/lzinga/treasury-mcp

If you end up using it I would love to hear about it. Feel free to open an issue on GitHub if you run into anything or have suggestions. I am still actively working on it and hope to add more data sources over time. If you know of any free or publicly available government APIs that aren't covered yet please let me know. The more sources that are available the more robust the cross referencing can be and the easier it is to get a fuller picture of what is actually going on.

I would encourage anyone interested to give it a try, even if you just want to ask your AI assistant a question about the national debt or what your state spends on healthcare. The data is there and it should be easy to get to.