Acquisition.gov MCP by 1102tools

npx add-mcp 'https://acquisition-gov.1102tools.com/mcp'

Acquisition.gov MCP

Free, open-source, read-only, deterministic MCP access to the official Acquisition.gov FAR Overhaul (RFO) model-part pages, the posted agency-deviation index, official deviation PDFs, and a small allowlist of RFO guidance resources.

Overview, demo video and hosted setup: https://1102tools.com/mcp/acquisition-gov

This server reports source content and metadata. It does not decide which rule governs a procurement. In particular, model deviation text is not treated as operative for an agency without that agency's posted deviation.

Local or hosted

Local (desktop)	Claude (hosted)	ChatGPT (hosted)
Local setup	Install	Install

• Local runs it on your computer, inside the Claude or ChatGPT desktop app or another AI app. You get your own full rate limits, and it relies only on the government service. It doesn't update itself, so ask your AI to update it. You don't have to set it up by hand: give your AI this page's link and ask it to set it up or walk you through it.
• Hosted is always up to date, needs no setup or keys, and works in Claude or ChatGPT anywhere.

Compare local and hosted · Matching prompts

Install

uvx --refresh-package acquisition-gov-mcp --from acquisition-gov-mcp acquisition-gov-mcp

The server uses stdio and requires no credentials.

Tools

Tool	Purpose
list_rfo_parts(part?, agency?, updated_since?)	List RFO model parts with official source dates and matching posted-deviation counts.
get_rfo_part(part, section?, cursor?, max_characters?)	Retrieve parsed, paginated model text for one FAR part.
list_rfo_agency_deviations(agency?, part?, limit?)	Discover posted deviation documents. At least one filter is required.
get_rfo_agency_deviation(source_id, page_start?, page_end?)	Resolve only an indexed source ID and return page-numbered PDF text and document-found applicability language.
get_rfo_guidance(resource, heading?, cursor?)	Retrieve the FAQ, policy-and-guidance page, or FAR Council deviation-guidance PDF.

Every retrieved source includes a canonical URL, UTC retrieval time, SHA-256 content hash, extraction status, and warnings. Agency PDF dates and applicability are returned only when explicit labeled language or a standalone letterhead date is found inside the document; filenames are never used to infer them.

Safety boundary

• Only https://acquisition.gov and https://www.acquisition.gov are allowed.
• Redirect targets are revalidated; credentials, explicit ports, arbitrary hosts, and private IP targets are rejected.
• Downloads are bounded to 5 MiB HTML / 25 MiB PDF. HTML complexity, redirects, page selection and output length are also bounded.
• PDF extraction runs in one isolated parser subprocess at a time, with a 30-second deadline, 2 MiB decoded page-stream limit, and 160 MiB address-space / 8-second CPU limits on Linux. macOS and Windows retain process isolation and explicit content/output limits without the Linux resource caps.
• HTTP 429 is not burst-retried. Retry-After is retained in the shared pacing state.
• If the Python TLS transport stalls against Acquisition.gov's CDN, the server may use an installed system curl for the same prevalidated URL; redirects remain disabled and revalidated by the server.
• Duplicate and conflicting index entries are returned with warnings instead of silently resolved.
• Scanned, encrypted, and malformed PDFs return explicit extraction status and metadata where possible.

Regression tests, covering parser isolation, cancellation, real stdio and HTTP checks, and each tool's answers. See testing.md.