Introduction
Citehouse is the white-label retrieval-and-payments gateway that lets engineering societies sell per-call agent access to their licensed corpus. Builder customers integrate one MCP endpoint and receive citation-grade retrieval over every society on the gateway. Societies retain rights, brand, and customer-approval authority. Citehouse hosts the indexing, the retrieval, the citation logic, and the metering.
What the gateway does
The gateway handles three tasks in one round trip:
- Authenticate and meter the call. Every call is attributed to a customer and a society. The ledger entry is written before the response leaves the gateway.
- Retrieve from the licensed corpus. Top-K matched sections, ranked by a society-tuned reranker. Results are character-for-character faithful to the source standard.
- Cite the passage. Every passage carries a canonical citation in the society's preferred format, e.g. ASME B31.3 §304.1.2.
What it does not do
- The gateway does not store or transmit your end-user's identity beyond what the calling agent sends.
- The gateway does not return synthesized text. Passages are excerpts, not summaries.
- The gateway does not perform engineering judgment. Application correctness is the calling product's responsibility.
Gateway endpoints
The gateway exposes both an MCP server and an HTTP API.
MCP
The MCP server lives at mcp.citehouse.com. It exposes four tools:
query_standard— natural-language retrieval against an indexed standard.lookup_section— direct fetch of a specific section by canonical identifier.cite_passage— given a retrieved passage and bibliographic metadata, return the canonical citation in the society's preferred format.list_corpora— discovery: returns the list of corpora the calling agent has access to.
HTTP
REST endpoints under /api/v1/:
GET /api/v1/corpora— list corpora the caller can access.POST /api/v1/corpora/:id/query— query a specific corpus.GET /api/v1/standards/:standard_id/sections/:section_id— direct section lookup.GET /api/v1/calls/:id— fetch a metering record.
x402-compatible endpoints respond 402 Payment Required on first call without an authenticated wallet token; the client retries with a settlement token. See x402 integration.
Performance contract
| Metric | Target |
| --- | --- |
| p95 latency | 800 ms |
| p99 latency | 1.4 s |
| Citation correctness | published per-corpus at /standards/[society]/eval |
| Availability | 99.5% production SLA |
Latency targets exclude the LLM rerank/synthesis path. Calls that opt into reranking add 200–600 ms.
Next steps
- MCP integration walks the TypeScript MCP path end to end.
- x402 integration describes the payment-required handshake.
- Citation format documents the canonical citation block.