citature MCP Server — cited federal money data as agent tools

A Model Context Protocol server that gives an AI agent direct, tool-shaped access to citature's cited federal data: campaign contributions, contract awards, follow-the-money flows, and committee/candidate lookup. Every tool returns a receipt — the exact public source URL, when it was retrieved, coverage caveats, match provenance, and guardrails — so an agent can verify every number against the primary federal record.

Runs in Apify Standby mode: the server stays warm and answers over Streamable HTTP at /mcp, so tool calls are fast and the underlying data connection is established once, not per request.

Keywords: MCP server, Model Context Protocol, AI agent tools, campaign finance API, federal spending API, FEC data, USAspending, government contracts, follow the money, cited data, agent-ready, structured tool output.

Tools

Each tool takes a small typed input and returns both human-readable text (the answer) and structuredContent (the full receipt).

Fourteen tools, on two pricing tiers — the eleven standard tools and the three govcon tools (marked below), which meter higher for parity with the standalone govcon actors (see Pricing).

Tool	Input	Returns	Tier
contributions_by_state	cycle, limit?	contribution totals by contributor state	standard
contributions_by_committee_type	cycle, limit?	contribution totals by committee type	standard
awards_by_agency	fiscal_year, limit?	award totals by awarding agency	standard
awards_by_state	fiscal_year, limit?	award totals by place-of-performance state	standard
awards_by_recipient	fiscal_year, limit?	award totals by recipient (UEI)	standard
follow_the_money	committee_id, cycle?, limit?	contribution inflows by contributor state	standard
committee_lookup	committee_id	one committee (or null)	standard
candidate_lookup	candidate_id	one candidate (or null)	standard
committee_search	name_prefix, limit?	committees by name prefix	standard
candidate_search	name_prefix, limit?	candidates by name prefix	standard
recompetes_expiring	expiring_from, expiring_to, naics?, agency?, limit?	federal contracts expiring in a YYYY-MM window, soonest first, each with the incumbent (by UEI) and its obligation	govcon
incumbent_history	uei? or recipient? (one required), agency?, naics?, limit?	a contractor's contract lineage per agency and NAICS: count, total obligated, fiscal-year span, latest period-of-performance end	govcon
agency_buying	agency?, naics?, fiscal_year?, limit?	federal obligations, contract counts, and distinct-contractor counts by awarding agency, NAICS, and fiscal year	govcon
opportunity_search	naics?, agency?, posted_from?, posted_to?, active?, limit?	open and forecasted procurement opportunities across SAM.gov and Grants.gov, newest first (no standalone actor)	standard

Every tool description embeds the guardrail caveats below, so an agent reads them before it calls.

Connecting

Point any MCP client at the Standby endpoint (Streamable HTTP transport):

https://<your-actor-standby-url>/mcp

A tool call returns a receipt as structuredContent. For example, calling awards_by_agency with { "fiscal_year": 2024, "limit": 2 }:

{
  "answer": [
    {
      "fiscal_year": 2024,
      "agency_entity_id": "agcy_fbc1…",
      "agency_name": "Department of Defense",
      "total_cents": 5647517724781,
      "award_count": 113
    }
  ],
  "citations": [
    {
      "source": "usaspending",
      "source_label": "USASpending.gov",
      "query_url": "https://api.usaspending.gov/api/v2/search/spending_by_award/?page=1",
      "retrieved_at": "2026-01-02T00:00:00.000Z",
      "as_of": null,
      "cache": "cached",
      "record_id": null
    }
  ],
  "coverage": ["Coverage reflects records currently in citature's dataset, not the full upstream history."],
  "guardrails": [
    "Research over public-domain federal records — not legal, financial, or eligibility advice.",
    "citature reports what the record says. It never renders a compliance verdict, a screening pass/fail, or a recommendation.",
    "Entity-level records only (committees, candidates, recipients, agencies, contracts). No individual-person data."
  ],
  "matches": [
    {
      "match_method": "exact",
      "match_confidence": 1,
      "left_key": "usaspending.awarding_agency",
      "right_key": "citature.entity_agency.agency_name",
      "note": "agency resolved to a citature entity on the toptier awarding-agency name"
    }
  ],
  "schema_version": "0.1.0"
}

Money is always integer cents. retrieved_at is when citature fetched the record; a source's own as-of date, when it publishes one, is a separate field.

Guardrails

• Research over public-domain federal records — not legal, financial, or eligibility advice.
• citature reports what the record says. It never renders a compliance verdict, a screening pass/fail, or a recommendation.
• Entity-level records only. No individual-person data is ever returned.

Pricing

Pay per tool call, on two tiers. The eleven standard tools meter on the tool-call event; the three govcon tools — recompetes_expiring, incumbent_history, and agency_buying — meter on the higher tool-call-govcon event, so this MCP surface never undercuts the standalone govcon actors. A call bills at the rate of the tool it invokes. The per-event prices are set in the Apify Console, which is the authoritative record of what a call is billed.

You are charged only when a tool returns an answer; a rejected (invalid input) or failed call is never billed.