Use this index in your own code or agent
Every agent and vendor entry on this site is also available as structured data: a bulk JSON feed, a per-entity endpoint, and an MCP server for an AI assistant to query directly. All three return the same evidence — source URL, retrieval date and verbatim quote — that the HTML page shows, licensed CC BY 4.0. As of now, that's 771 agents and 35 assurance vendors.
Bulk JSON
/api/index.json returns every published entity (name, slug, type, subtype, developer, homepage, summary, canonical URL) and every published article, plus a link to each entity’s own per-field endpoint. Cached 5 minutes at the edge.
curl https://aidispatch.news/api/index.jsonPer-entity JSON
/api/entities/[slug]/ returns one entry in full: every field grouped by category, each carrying its value, status (present, absent, or not yet interpreted), the verbatim quote, the source URL, the retrieval date, and the entry’s trust-gap scores with their rationale. The slug is the same one in the entry’s own page URL.
curl https://aidispatch.news/api/entities/armilla-ai/MCP server
https://aidispatch.news/api/mcp/ speaks the Model Context Protocol over stateless Streamable HTTP: POST JSON-RPC, no session, no API key. Four tools are exposed:
| search_index | Search published entries by name, summary or developer; filter by entity type or subtype. |
| get_entity | Full record for one entry by slug, with every field's quote, source and retrieval date. |
| get_trust_gaps | Entries at or below a score threshold on one of the five rubric dimensions, with the rationale. |
| list_categories | Every published category with entry counts, split by agent vs. vendor. |
A generic MCP client only needs the URL above; most will discover the tools themselves via tools/list. For a direct call:
curl https://aidispatch.news/api/mcp/ \
-H 'Content-Type: application/json' \
-d '{
"jsonrpc": "2.0",
"id": 1,
"method": "tools/call",
"params": { "name": "search_index", "arguments": { "query": "insurance" } }
}'The server’s own initialize response states the evidence standard and attribution terms before a model sees any data — the same rules as this page, just addressed to the tool caller rather than the reader. See /llms.txt for the equivalent brief aimed at a model deciding whether to cite this site at all.
Frequently asked
- What's the licence?
- CC BY 4.0. You may copy, redistribute, remix and build on the data, including commercially, as long as you attribute AI Dispatch and link back to the entry or query you used. The same terms apply to the bulk JSON, the per-entity JSON, and the MCP server — there is one licence, not three.
- What attribution is required?
- Name AI Dispatch and link to https://aidispatch.news, or to the specific entry URL returned with the data you used (each entity, article and tool result carries its own canonical URL for exactly this). There's no required wording beyond that.
- Is the data rate-limited or does it need an API key?
- No key. The bulk and per-entity JSON endpoints are plain cached GETs. The MCP server is unauthenticated too, since everything it returns is already published under the same licence — it caps request bodies at 64KB and a shared instance at 240 calls per minute as a circuit breaker against a runaway loop, not a quota you need to request access to.
- How is this different from scraping the HTML pages?
- Every field returned by these endpoints carries the same source URL, retrieval date and verbatim quote shown on the page it came from — the structured form of exactly what's on the page, not a separate or richer copy. Where we have no information, the field says so explicitly ('no public information') rather than being omitted, so an absence in the HTML and an absence in the API always agree.
- Can I trust a value without re-checking the source myself?
- Every present value was checked against its source by a database trigger before it could be written: a quote that doesn't appear verbatim in the stored source document is rejected on insert. You're still welcome to re-check it — the source URL is right there — but the quote itself cannot be fabricated by construction, not just by editorial policy.
How the data is verified
Every field above cites a source document and a retrieval date, and a quote cannot be stored unless it appears verbatim in that document — enforced by a database trigger, not editorial policy alone. See the full methodology for the rubric behind the trust-gap scores and how absences are handled.