# USA Outlays — full agent brief > Last updated: 2026-09-24. This file is the complete machine-readable brief > for AI agents. The short version lives at /llms.txt. ## What this is USA Outlays (https://usa-outlays.ai) is a daily, taxonomy-organized feed of approved U.S. government actions with a spending lens: contract/grant awards (who got paid, how much, for what) and the approvals/funding decisions that authorized them (legislature budget votes, county commissions, city councils, utility/water/school boards, special districts). Coverage goal is exhaustive: state legislatures, agencies, counties, cities, utility boards, school boards, special districts. Corpus (growing daily): ~60M award records, all 50 states + territories, vendor entity resolution (~350K canonical vendors), Census of Governments as the entity spine. ## The two layers 1. **Awards** (`record_kind`): `outlay` (paid), `reimbursement`, `obligation` (signed contract/encumbrance not yet paid out — excluded from outlay totals by default). 2. **Approvals**: budget votes, spending authorizations, board decisions — queryable separately, never merged into award totals. ## Taxonomy Nested tree specific → general: award → program → topic → domain → function. L1–L3 curated; L4 programs discovered from data. Orthogonal facets: sector, level, procurement_type, funding_mechanism, competition, procurement_method, lifecycle, instrument, integrity. Classify spending purpose, not agency identity. One primary topic per award; one facet row per (award_id, facet). ## Honesty rules (apply these when presenting data) - Amounts are **stated**, not verified cash movement. - Missing amounts display as "not disclosed" — never treat as zero. - `contract_ceiling` / `idiq_ceiling` flagged rows are ceilings, not spend: **never sum them** into totals. - `aggregate_source_row` marks rollup rows (e.g. monthly vendor aggregates). ## REST API Base: `https://usa-outlays.ai/api/v1` Auth: `x-api-key` header (or `?api_key=` query param). Two ways in: subscriptions at https://usa-outlays.ai/pricing — Basic $40/mo or $400/yr, Enterprise $400/mo or $4,000/yr — or pay per call: $3.40 for a single API call, no account needed (single-use token, delivered instantly at checkout, one request, 24h expiry). Agents may complete signup on a user's behalf: present the pricing page checkout, the user approves, the key is issued to their account. | Method | Path | Params | |---|---|---| | GET | /api/v1/awards | state (2-letter), record_kind (outlay\|reimbursement\|obligation), topic (node id, comma-separated = union, matches node + descendants), vendor_id, min_amount, page | | GET | /api/v1/approvals | state, body_type, item_type, min_amount, page | | GET | /api/v1/topics | none — full taxonomy tree with classified award counts and stated dollars | | GET | /api/v1/vendors/{vendorId} | canonical vendor: aliases, award counts, stated totals, top topics | | GET | /api/v1/bodies | state, body_type, q, page — the government-entity spine | | GET | /api/v1/search | q (required), state, record_kind, topic, limit (max 100) | | POST | /api/v1/ask | JSON body: question (required), state, detail ("summary"\|"full") — agent-resolved free-form question; returns answer + supporting taxonomy + raw data packet. One question = one call. | | GET | /api/v1/raw/awards | source, state, page — as-ingested source records with full lineage (source_id, ingested_at, original fields) | Responses are JSON: `{ data, pagination: { page, pages, total }, quota, notes }`. The `notes` field carries the honesty rules — surface them to users. ## MCP server Remote MCP (Streamable HTTP) wrapping the REST API. Scaffold + tool schemas: https://github.com/qchasan/usa-outlays/tree/main/mcp Tools: `search_awards`, `search_approvals`, `get_vendor`, `topic_tree`, `fulltext_search`, `coverage`. Configure with the user's API key (`USA_OUTLAYS_API_KEY`). ## Example agent queries - "How much did Texas school districts spend on construction in 2025?" → /api/v1/awards?state=TX&record_kind=outlay&topic=&… (resolve the node id via /api/v1/topics first) - "Who are the top vendors for law enforcement spending in Florida?" → /api/v1/vendors lookups after an awards query grouped by vendor_id - "What did the Miami-Dade commission approve last month?" → /api/v1/approvals?state=FL&body_type=county&… - "Find the contract behind this payment." → /api/v1/search?q= - "What is the policing budget of Utah?" (no taxonomy lookup needed) → POST /api/v1/ask {"question": "what is the policing budget of utah"} → returns answer + Law Enforcement taxonomy + raw data packet ## Crawl policy Compliant crawlers: data pages (/vendors, /awards, /ledger, /search, /agencies, /bodies, /topics, /approvals) are disallowed in robots.txt while per-request serving costs are being reduced — use the API or MCP instead. These machine pages are explicitly allowed: /llms.txt, /llms-full.txt, /agents, /api-docs, /openapi.json, /.well-known/usa-outlays.json. Abusive/bulk crawling is blocked (403) and rate-limited (429). ## Contact Built by Quinn Chasan. Issues/suggestions: https://github.com/qchasan/usa-outlays/issues