Install Claude Desktop. Double-click to install.

Eight federal data and policy APIs ship as Claude Desktop Extensions (.mcpb). Download the file, double-click, done. No Terminal. No config editing. No uv install. Claude Desktop handles the runtime and prompts for API keys during install.

What you're installing
Deterministic. Each MCP server runs tested Python. Claude doesn't write fresh API-call code on every turn. Same input always returns the same structured output.
Production-hardened. 719 regression tests across 8 MCPs. 3-6 audit rounds per server. ~350 bugs fixed during hardening. Each server ships a testing.md.
One-click or manual. .mcpb gives you double-click install in Claude Desktop. Copy Config gives you the JSON for any local MCP client (Cursor, Cline, Zed, Claude Code, Continue) with no install warning.
Low context cost. Tool schemas are ~100 tokens each. All 8 MCPs combined cost less context than a single legacy API skill.
1
Install Claude Desktop
Download and install Claude Desktop (Mac or Windows). Free. Same way you would install any normal app. A Claude Pro subscription is recommended for extended workflows; Free tier works for single-query lookups.
2
Register free API keys
Three free keys cover the data MCPs that need them. BLS (500 queries/day), api.data.gov (1,000/hr, covers Per Diem and Regulations.gov), SAM.gov (rotates every 90 days). USASpending, GSA CALC+, eCFR, and Federal Register need no key. Paste each key when the extension prompts during install.
3
Pick an install path on the Tools page
Visit /tools. Two install options per MCP: Copy Config (recommended — clipboard the JSON, paste into your client's config, no warning) or .mcpb (one-click for Claude Desktop, triggers the standard third-party extension warning). Same code either way.
One-click alternative

.mcpb (Claude Desktop)

Download and double-click. Triggers Anthropic's standard third-party extension warning: "Installing will grant this extension access to everything on your computer." Not specific to these MCPs — appears on every unverified third-party MCPB until directory review.

Walk-through
Orchestration skills (SOW/PWS, IGCE, OT)
These are still skills.Scope decisions, cost buildup, FAR-compliant methodology narratives. MCPs cannot express a decision tree; skills can. These call the MCP servers above for their data.
FreeSingle-skill lookups only.
ProLonger workflows without truncation.
SonnetFine for single-skill queries.
OpusUse for orchestration skills (SOW/PWS, IGCE).
Enable in Claude settings:Generate memory from chat history so keys persist. Cloud code execution and file creation so the .docx and .xlsx generators run.
Troubleshooting
Double-click does nothing?Make sure Claude Desktop is installed and has been opened at least once. .mcpb files are only recognized by Claude Desktop.
Tools not appearing in chat?Open Claude Desktop Settings → Extensions. Confirm the extension is enabled. Check for a red error badge.
SAM.gov key rejected?SAM.gov keys expire every 90 days. Regenerate at sam.gov, open the extension settings, paste the new key.
Agency network blocks API?Many agency networks block outbound calls to api.gov domains. Test from a personal connection first. File an ITSM ticket for allow-listing.
Want to skip the install warning entirely?Use Copy Config on the /tools page. Clipboards a JSON snippet, paste into Claude Desktop's claude_desktop_config.json (or any MCP client's config). Zero warning, same code as the .mcpb bundle.

What these MCPs do — and don't

Two install paths, same code. .mcpb is one-click and triggers Anthropic's standard third-party warning. Copy Config is manual and skips it entirely. Here's what the code actually does on your machine — auditable at github.com/1102tools/federal-contracting-mcps.

These MCPs DO
  • Make HTTPS calls to the specific federal APIs listed in each server's readme (SAM.gov, BLS, GSA, USASpending, api.data.gov, eCFR, Federal Register, Regulations.gov). Nothing else.
  • Read API keys from environment variables Claude Desktop sets at install time.
  • Return structured query results back to Claude for you to reason over.
  • Log errors to stderr (visible in Claude Desktop's extension debug panel).
  • Ship with full source on GitHub, a testing.md audit record per server, and MIT license for modification.
These MCPs DO NOT
  • Read any files on your computer. The servers have no file I/O to your disk.
  • Access your chat history, memory, or conversation context outside the tool arguments Claude passes in.
  • Send telemetry, usage data, or analytics anywhere. No third-party calls other than the federal APIs above.
  • Store your queries, API keys, or responses anywhere outside your local machine and Claude Desktop's config.
  • Modify any system settings, install software, or persist beyond the Claude Desktop process that spawned them.