Relevance AI scores 77.0 on the Kin Score, exemplar band, down 1.5 from 78.5. Its Agent Readiness is 65.3, and it reaches agent-native with no gate holding it back. For a company whose own description says it builds platforms for AI agents, that is the result you would hope for. What is less expected is where the score comes from, and the one facet that pulls against it.
Relevance AI describes a no-code builder for specialist agents and multi-agent “workforces,” a built-in evals system, 1,000+ integrations, and three ways in for developers: a REST API it puts at 566 operations on regional hosts in the US, EU and Australia, a JavaScript/TypeScript SDK, and a hosted, OAuth-protected MCP server at mcp.relevanceai.com.
What is in the surface
The provider record counts 2 published APIs: the MCP server and the REST surface. That REST surface is split on apis.io into 32 API pages, from Agents and Chat to Evals, Knowledge, Syncs and User Memory, plus one AsyncAPI event specification. The score’s provenance block counts 31 contracts, 0 derived, so these are Relevance AI’s own. Its callable share is 0.0. Every base URL reads https://api-{region}.stack.tryrelevance.com/latest, which leaves the region for the caller to fill in.
| Facet | Score |
|---|---|
| Access clarity | 92.1 |
| Discoverability | 75.0 |
| Developer ergonomics | 71.4 |
| Operational transparency | 65.8 |
| Contract quality | 52.5 |
| Contract governance | 18.2 |
Access clarity is the strongest facet: freemium, self-serve signup, a trial, apiKey and OAuth schemes, and a trust center that claims SOC 2 Type II and GDPR. Contract governance is the weakest at 18.2. Thirty-one contracts with no lifecycle signal behind them tell an integrator what exists today and nothing about what will change.
The agent record is the real story
Most of the agent row is lit, and lit by Relevance AI. Provenance marks the MCP server, the Agent Skills and the conformance artifacts first-party. Protected resource metadata is verified, idempotency is verified, delegated identity and auth clarity are served, dynamic client registration works, and the agent card is conformant. Few providers in the catalog have all of the OAuth discovery chain an MCP client needs actually responding. Relevance AI does.
The gaps are specific. There are no OpenAPI examples, no dry-run mode, and no consent identity, and the agentic-access dimension is unlit. The scopes artifact reads “OAuth 2.0 · no documented scopes,” which is a strange thing to find on a platform whose enterprise pitch is RBAC and human-in-the-loop approvals. An agent can authenticate, but it cannot ask for less than everything.
One fairness note: the coverage disclosure shows 24.0 of 61.0 catalog points earned first-party. That is a floor, not the full share, but it does mean part of this record is our work rather than theirs.
Takeaway
Three changes would move the number. Name concrete regional servers in the contracts so a client can call them without templating. Publish deprecation and versioning signals so contract governance has something to measure. And document OAuth scopes that match the RBAC the platform already enforces. Relevance AI has done the hard agent work. The plain contract work is what is left.
See the full profile at apis.io/providers/relevance-ai/.