Cerbos alternatives
The highest-scoring alternative to Cerbos is Composio at 97/100 on the AgentReady, against 46/100 for Cerbos.
All scored in Agent Auth & Permissions.
| Score | Alternative | Compare |
|---|---|---|
| 97 | Composio | Cerbos vs Composio |
| 79 | Permit.io | Cerbos vs Permit.io |
| 63 | Auth0 | Cerbos vs Auth0 |
| 61 | Arcade | Cerbos vs Arcade |
| 55 | Ory | Cerbos vs Ory |
| 51 | Stytch | Cerbos vs Stytch |
What Composio does that Cerbos does not
Composio passes clear canonical domain, structured api reference, openapi / spec quality, authentication documented, pricing understandable, request examples provided, response examples provided, errors and status codes documented, limits / constraints documented, self-service signup, no mandatory sales call, agent-compatible signup flow, programmatic credential creation, free trial or free allowance, fast time to first request, cli available, structured, predictable output, machine-readable errors, rate-limit behavior predictable, agent compatibility verified, and Cerbos fails them. That is the whole of the gap between the two scores, so if none of those matter to your agent the ranking above overstates the difference.
Why people look for an alternative
Cerbos fails clear canonical domain, structured api reference, openapi / spec quality, authentication documented, pricing understandable, request examples provided, response examples provided, errors and status codes documented, limits / constraints documented, self-service signup, no mandatory sales call, agent-compatible signup flow, programmatic credential creation, free trial or free allowance, fast time to first request, cli available, structured, predictable output, machine-readable errors, retry behavior documented, idempotency support, rate-limit behavior predictable, agent compatibility verified. Each one is a place an unattended run stops, which is usually what sends someone looking in the first place.
Where Cerbos loses the points
| Pillar | Cerbos | What it measures |
|---|---|---|
| Discover | 93 | Can an agent find it at all: sitemap, robots, structured data. |
| Understand | 15 | Is there a spec, an llms.txt, a machine-readable description. |
| Adopt | 40 | Signup, keys, SDKs, and whether a human has to be involved. |
| Operate | 35 | Rate-limit headers, errors, status, and behaviour under load. |
Cerbos is weakest on understand at 15/100. Without a spec an agent has to guess at endpoints, and it guesses wrong. If that is the part your agent depends on, take the ranking above seriously; if it is not, Cerbos may still be the right pick.
What an agent runs into with Cerbos
These are the findings from the last audit on 2026-08-23. Each one is a place unattended software stops, which is the practical reason to look at the list above.
- Make signup self-serve; an agent cannot sit through a sales call.
- Publish an OpenAPI spec so agents can generate a client.
- Flesh out the OpenAPI: paths, schemas, and descriptions.
- Document error responses as an enum in your OpenAPI.
- Let agents create API keys programmatically.
- Return X-RateLimit / Retry-After headers so agents pace themselves.
- Add request examples to your API reference.
- Add response examples to your API reference.
What this category actually supports
Across the 7 products scored in Agent Auth & Permissions, delegate scoped access is offered by 14%. Anything under half is a gap you should expect to fill yourself whichever one you pick.
An agent can get this list directly: resolve("alternative to Cerbos") over the MCP server, or GET /v1/categories/agent-auth.