Copilot cloud-agent administrators can copy the allowlisted configuration from the public distribution repository into their repository MCP settings.
02
Endpoint matrix
Every endpoint speaks the same seven-tool server contract and the same free-beta policy. A client or marketplace manifest may intentionally allowlist a smaller subset. Separate paths provide privacy-safe aggregate channel attribution; they do not identify an individual caller.
Distribution channels, their endpoints, and how many tools each manifest advertises.
Channel
Endpoint
Advertised
Canonical
https://grants.utilityhouse.xyz/mcp
Any standard Streamable HTTP client. All seven tools.
Claude
https://grants.utilityhouse.xyz/mcp/claude
All seven tools.
OpenAI / ChatGPT
https://grants.utilityhouse.xyz/mcp/openai
The current review package declares all seven tools.
Official MCP Registry
https://grants.utilityhouse.xyz/mcp/registry
All seven tools. Non-payment surface.
Smithery
https://grants.utilityhouse.xyz/mcp/smithery
All seven tools. Non-payment surface.
Cursor
https://grants.utilityhouse.xyz/mcp/cursor
All seven tools.
Gemini
https://grants.utilityhouse.xyz/mcp/gemini
Manifest allowlists four tools.
GitHub Copilot
https://grants.utilityhouse.xyz/mcp/copilot
Manifest allowlists four tools.
x402 (future paid)
https://grants.utilityhouse.xyz/mcp/x402
Separately governed. Free during the current beta; a paid launch requires its own release.
Apify is a separate three-operation read-only subset. Directory endpoints remain non-payment surfaces even after a separately approved paid launch.
03
Tool contract
Seven advertised tools. Legacy names remain callable for compatibility but are not advertised.
search_federal_grantsfree
Use when
A user needs current federal opportunities from an official public source.
Input
A search query and an optional result limit.
Output
Normalized opportunity identifiers, titles, agencies, status, dates, and official URLs.
Behaviour
Read-only against the public opportunity source. A recent public-source response may be reused for up to six hours; the result labels live, fresh-cache or stale-fallback.
analyze_federal_grantfree
Use when
One opportunity is known and the user wants a preview of supplied organization facts against official hard gates.
Input
Opportunity ID plus a minimal profile: entity type, state, optional country, optional requested amount, optional cost-share capacity. No names, identifiers, budgets or narratives.
Output
Recommendation, reason codes, hard-gate evidence, funding fields, history availability, unresolved states, sources, limitations, and chargeable: false.
Behaviour
Never returns a confident GO. A missing source produces UNKNOWN, not a pass.
get_federal_grant_evidence_packlimited free beta
Use when
The user wants a versioned hard-gate and award-history pack for one opportunity.
Input
A stable request_id, opportunity ID, entity type, state, requested amount, and cost-share capacity.
Output
A versioned complete-result pack with its digest, or a non-chargeable unavailable result.
Behaviour
The server completes and records the result before release. If any included hard gate or Assistance Listing snapshot is incomplete it returns unavailable and does not consume beta capacity. Must be sent as a single JSON-RPC message, not inside a batch. Inspect the public result sample and the live data status before integrating.
Opportunity 359701 only, when the user wants to test current actual non-personal organization, project, and readiness facts.
Input
Enumerated non-personal facts, plus separate current-facts attestation, intent to apply, and 30-day storage consent.
Output
A derived pointer-only result and an attestation receipt.
Behaviour
The raw profile, names, contact details, identifier values, narratives, documents and credentials are never stored. Missing facts stay visible rather than being guessed. Attestation is not independent verification.
get_grant_document_manifestfree beta
Use when
An agent needs to know which official attachments exist for an opportunity and whether that set is complete.
Input
An opportunity ID.
Output
The official attachment set with a completeness state.
Behaviour
Missing, quarantined or stale evidence remains unresolved rather than being omitted silently.
analyze_nofo_requirementsfree beta
Use when
An agent needs pre-award requirement evidence located inside the notice.
Input
An opportunity ID and the requirement scope.
Output
Source-located pre-award requirement evidence.
Behaviour
Requirements keep the document locator they were read from. Absence of evidence is reported as absence, not as 'not required'.
compare_grant_amendmentsfree beta
Use when
An agent needs to know what changed between two revisions of one opportunity.
Input
An opportunity ID and the revisions to compare.
Output
Reported revision changes.
Behaviour
Reports a change only when both sides have inspectable official evidence; otherwise the comparison stays unresolved.
04
Limitations
What the contract does not cover, and how the server behaves when evidence is missing.
Declared limitation
The complete pack covers cited hard gates and versioned award history. It does not evaluate semantic project relevance, determine legal eligibility, predict an award, write an application, or submit one.
Advertised tools do not change Grants.gov or another external service. The opt-in proof tool writes only the consented derived result and receipt.
Each valid tool call increments one aggregate daily counter for its distribution channel, tool, and success or error outcome. No caller identity or request content is stored in that usage table.
Missing, stale, unavailable, and capacity-limited states fail closed and never request payment.
A future paid release is planned at $0.09 per complete result; it is not enabled during this beta.