Agents act on entity facts — who founded a company, when, who owns it, whether it still exists. Those are exactly the facts AI engines confabulate most confidently, and a hallucinated fact becomes a failed action in production. Entidex maintains resolved entity records with provenance on every observation, signs them, and exposes them where machines look: an MCP server, a public API, per-entity Knowledge Statements, and llms.txt.
One layer. Any agent. Verified.
Connect over the hosted MCP endpoint — nothing to install or self-host — or call the REST API directly. Full MCP client config (Claude Desktop, Cursor, and other hosts) is on the MCP server page.
{
"mcpServers": {
"entidex": {
"url": "https://entidex.com/api/mcp",
"headers": { "Authorization": "Bearer entx_live_sk_..." }
}
}
}Grounding the input context is half the job; the other half is checking what your agent is about to say. Grounding Check takes a generated statement about an entity and returns a coded verdict per assertion — supported, contradicted, or unverifiable— with confidence and a citation into the signed verified record. It is a deterministic check (no model judges truth): granularity differences like “Austin” vs the verified “Austin, Texas” come back supported, and assertions the record doesn’t cover come back unverifiable — nothing is guessed. Free tier, rate-limited.
# After your model drafts an entity claim, check it BEFORE returning it.
curl -s -X POST "https://entidex.com/api/v1/entities/tesla/grounding-check" \
-H "Authorization: Bearer entx_live_sk_..." \
-H "Content-Type: application/json" \
-d '{"statement": "Tesla is headquartered in Palo Alto and was founded in 2003."}'
# → verdict: "contradicted" (HQ conflicts with the verified record;
# "founded in 2003" is supported). On contradicted: swap in the
# assertion's citation.verifiedValue and cite the source + date.
# → verdict: "unverifiable": the record doesn't cover the assertion —
# qualify the claim or drop it; the check never guesses.entity_resolve — resolve a raw string to a canonical Entidex ID with a confidence band. (Starter+)entity_verify_fact — verify facts against the verified record: verified values with source and date, the correction delta where AI engines diverge, and signed provenance. (Starter+)entity_grounding_check — check a statement before asserting it: coded verdicts (supported / contradicted / unverifiable) with citations against the signed verified record. (Free)entity_intelligence — the layered intelligence view (profile + scores) for a resolved entity. (Pro+)entity_drift — drift history, named events, and lead-lag across surfaces. (Pro+)entity_decision_survival — latest decision-survival result with the displacement chain. (Pro+)entity_knowledge_analysis — what AI engines get wrong, miss, or disagree on. Compact summary by default; use section or detail for per-field data. (Pro+)entity_submit_feedback — submit a correction back to Entidex. Captured append-only with provenance and reviewed before it affects the published record. (Starter+)Each tool enforces the same access level as its REST equivalent.
Every entity carries a public, crawlable Knowledge Statement — verified facts plus corrections to common AI claims, available as Markdown, schema.org ClaimReview, or JSON, and cryptographically signed. For agents and CI, the authed /v1/entities/<id>/ground-truth endpoint returns the same record. Discovery starts at the site-root llms.txt, and a machine descriptor of the MCP server lives at /.well-known/mcp.json.
Free tier: resolution and Knowledge Statement reads, rate-limited. Developer API is £29/mo with monthly credits and pay-as-you-go overflow.
Create a free account to get an API key (1,000 monthly credits, 1,500 with a verified email + phone), then resolve, read and ground in minutes. Or try the keyless scan first.