Ask your AI assistant about federal contracts in plain English. This is a walkthrough of the GovCon API MCP server – an open-source bridge that exposes SAM.gov opportunities, awards, and exclusions to any MCP-compatible AI client.
A real conversation in Claude Desktop. One natural-language question, one tool call, a structured summary back. Nothing simulated.
What are the latest federal AI opportunities from the last week?
21 AI-related opportunities this week:
Want me to filter by agency, set-aside, or export to spreadsheet?
search_opportunities tool and summarizing results.Searching SAM.gov manually is painful. The UI is slow, agency names are stored inconsistently (the same agency appears under 20+ name variations), and there's no way to ask a clean question like “what contracts has Booz Allen won this quarter?” without clicking through dozens of filters.
This MCP server gives you the same SAM.gov data behind a natural-language interface, in the AI assistant you already use daily. $19/month. No dashboard to learn. No sales call. No annual contract. Install in 30 seconds, ask in plain English.
All SAM.gov opportunities, normalized: 59 structured fields per notice, descriptions expanded (SAM.gov's own API hides them), agency names mapped to canonical values, 200K+ records updated daily.
Six tools exposed: search, single-record fetch, incremental changes, agency lookup, exclusion screening, award history. You don't memorize endpoints – the AI reads the natural-language question and calls the right one.
Ten real prompts that work out of the box. Your AI picks the right tool automatically – you don't need to know it exists.
“Find all DoD cybersecurity solicitations posted this month over $1M.”
search_opportunities · naics=541511, keywords=cybersecurity, value_min=1000000, posted_after=2026-04-01
“Show me HUBZone-set-aside opportunities in NAICS 541330 from the last 60 days.”
search_opportunities · set_aside=HZC, naics=541330, posted_after=2026-02-13
“What contracts has Booz Allen Hamilton won in the past quarter?”
search_awards · awardee="Booz Allen", date_from=2026-01-01
“What's FEMA spending on this year? Use the right canonical name.”
lookup_agency("FEMA") → search_awards(agency="FEDERAL EMERGENCY MANAGEMENT")
“Has [vendor name] ever been on the SAM.gov exclusions list?”
check_exclusion · name="Acme Corp"
“Show me everything new in federal contracts since yesterday 9am.”
recent_changes · since=2026-04-12T09:00:00Z
“What's the biggest award in NAICS 236220 construction contracts this year?”
search_awards · naics=236220, date_from=2026-01-01, sort_by=value
“Get me the full details on notice ID W9126G-25-Q-0001 including contacts.”
get_opportunity · notice_id="W9126G-25-Q-0001"
“California SDVOSB IT opportunities, deadline in next 30 days.”
search_opportunities · state=CA, set_aside=SDVOSBC, naics=541511, due_before=2026-05-13
“Pull last 30 days of Navy sources-sought notices and format as a table.”
search_opportunities · agency="Navy", notice_type="Sources Sought", posted_after=...
Same install command everywhere; only the config path changes per client. Python 3.10+ required.
Installs to ~/.local/bin/govconapi-mcp, outside sandboxed folders like Documents.
14-day free trial at govconapi.com (no credit card). After trial: $19/mo Developer plan or $39/mo Pro. MCP is included with any paid plan.
Edit the config file:
~/Library/Application Support/Claude/claude_desktop_config.json%APPDATA%\Claude\claude_desktop_config.json{
"mcpServers": {
"govconapi": {
"command": "govconapi-mcp",
"env": {
"GOVCONAPI_KEY": "gca_your_key_here"
}
}
}
}
Quit and reopen Claude Desktop. The govconapi tools appear in the tool picker.
One command adds it for your user:
Or edit ~/.claude/settings.json manually using the same JSON as Claude Desktop.
Settings → Cursor Settings → MCP → Add new MCP server.
Or edit ~/.cursor/mcp.json directly:
{
"mcpServers": {
"govconapi": {
"command": "govconapi-mcp",
"env": { "GOVCONAPI_KEY": "gca_your_key" }
}
}
}
Install Continue in VS Code, then edit ~/.continue/config.yaml:
mcpServers:
- name: govconapi
command: govconapi-mcp
env:
GOVCONAPI_KEY: gca_your_key
Assistant panel → MCP Servers → Add Context Server. Or edit ~/.config/zed/settings.json:
{
"context_servers": {
"govconapi": {
"command": "govconapi-mcp",
"env": { "GOVCONAPI_KEY": "gca_your_key" }
}
}
}
Spawn the MCP server as a subprocess using the official mcp client SDK:
# pip install mcp from mcp import ClientSession, StdioServerParameters from mcp.client.stdio import stdio_client params = StdioServerParameters( command="govconapi-mcp", env={"GOVCONAPI_KEY": "gca_your_key"} ) async with stdio_client(params) as (read, write): async with ClientSession(read, write) as session: await session.initialize() tools = await session.list_tools() result = await session.call_tool( "search_opportunities", {"naics": "541330", "limit": 10} )
Works with any LangChain / LlamaIndex / custom agent that supports MCP tool adapters.
The govconapi-mcp package is a thin Python wrapper (~250 lines) around our REST API. Your AI client spawns it as a local subprocess over stdio. The subprocess reads GOVCONAPI_KEY from its environment, proxies each tool call to https://govconapi.com/api/v1/... over HTTPS, and returns structured JSON.
Because it wraps the REST API, everything else carries over automatically: your rate limits, plan restrictions, filter validation, and request logging. Usage shows up in your normal account dashboard.
Open source on PyPI under MIT. Audit it, fork it, run it locally with your own API key.
Everything the MCP server exposes. Your AI reads the prompt and picks the right one automatically.
| Tool | Purpose | Typical prompt |
|---|---|---|
| search_opportunities | Filtered search across 200K+ records | “Find AI solicitations over $1M” |
| get_opportunity | Fetch one notice, all 50+ fields | “Details on notice X” |
| recent_changes | Incremental sync since a timestamp | “What's new since yesterday” |
| lookup_agency | Map acronyms to canonical names | “What's the SAM.gov string for FEMA?” |
| check_exclusion | Debarment screening | “Is X excluded from federal contracts?” |
| search_awards | Find who won contracts | “What did company Y win last quarter?” |
Common questions. If something isn't here, email [email protected].
No. MCP access is included with the Developer ($19/mo) and Pro ($39/mo) plans. Same API key, same rate limits. The 14-day free trial also includes MCP (25 calls/day, no credit card).
Our 14-day trial is free (25 calls/day, no credit card). After that, $19/mo for the Developer plan covers MCP plus the full REST API.
ChatGPT's desktop app and Apps SDK both support MCP. Same install, same config format. You'll want an OpenAI plan that supports MCP (Enterprise, Teams, or the Apps SDK beta). For out-of-the-box use, Claude Desktop is the smoothest.
Yes, all of them. MCP is an open standard – any client that speaks it works identically. See the install section above for each client's config file path.
Same as our REST API: 25 calls/day on free trial, 1,000 calls/hour on Developer plan, with a 60/minute burst cap. Each tool call from your AI counts as one API call. You won't hit limits with normal interactive use.
Everything in GovCon API: 200K+ federal contract opportunities (solicitations, award notices, sources sought, presolicitations), 50K+ contract awards with dollar amounts and winners, 160K+ excluded (debarred) entities, and an agency crosswalk for name normalization. All sourced from SAM.gov, cleaned and updated daily.
Most common cause: the binary is inside a sandboxed folder (~/Documents). Claude Desktop can't spawn processes from there. Use pipx install govconapi-mcp instead of a manual venv – it puts the binary at ~/.local/bin/ which sandboxes can reach. Check the log at ~/Library/Logs/Claude/mcp-server-govconapi.log for details.
The key lives in your local client's config file. It's sent over HTTPS in the Authorization header on each call. The MCP server never logs, transmits, or exposes it anywhere else. Source code is public and auditable on PyPI.
Yes – that's the main reason we built it. Install Claude Desktop, paste one config block with your API key, and ask in plain English. No command line, no JSON parsing, no pagination management. Your AI handles all of it.
Model Context Protocol – an open standard from Anthropic that lets AI assistants call external tools. Think of it as “the API protocol for AI”. When you install an MCP server, your AI gains new capabilities. Ours gives it six new abilities specifically for federal contracts data.
Email [email protected]. The MCP server is open source – PRs welcome. We ship updates fast because it's a thin wrapper over the REST API.
Grab a 14-day trial key (no credit card), paste one JSON block into your Claude / Cursor / VS Code config, and ask your assistant a question about federal contracts. The same key works for the REST API too.