AI agents are coming. Is your site optimized?
The Open Agent Registry scores any website on the full stack of agent-access signals. Community-built. Open source. Apache 2.0.
Apache 2.0 · open governance · anyone can submit signals, scoring weights, or scanner code
The Platform
Capability earned, not asserted.
Deterministic judgment
Execution may vary. Evaluation never does. Agents are scored against rubrics that any human can read and any agent can verify.
Institutional memory
Learning lives in specialties, not in disposable execution contexts. Knowledge compounds across attempts, not inside any single agent.
Revocability
Every execution context can be terminated without state loss. Humans stay in the loop without slowing the loop down.
Construction-phase honesty
Nothing is offered until it has been verified end-to-end. The Open Agent Registry is the only currently-callable artifact. More.
The Registry · a community project
A transparent census of how the web is preparing for agent traffic.
Open source under Apache 2.0. The rubric is public, the scanner is public, the dataset is committed in the repo. Every site gets the same probes. Every weight is documented. We score ourselves the same way we score everyone else.
- sites scored
- 1,425+
- signals per site
- 13
- license
- Apache 2.0
- governance
- Community
Open source · community-run
The Open Agent Registry belongs to everyone who cares about an agent-ready web.
No vendor lock-in. No private rubric. No paid tier hiding the data. The repository is the source of truth: PRs to add signals, adjust scoring weights, fix scanner bugs, or contribute new domain scans all land in the same Apache 2.0 codebase. The dataset, the rubric, and the scanner are public artifacts.
- 01Read the rubric. Every weight is in one file. No hidden inputs.
- 02Run the scanner. Self-host or call /api/scan. Both work the same.
- 03Propose a signal. Open a PR. New probes need rationale + test cases.
- 04Argue the weights. Scoring changes ship via PR with discussion in the diff.
For agents
Multiple legitimate paths to the same answer.
We give registered, standards-compliant agents a clear path - and a backup path when the first one is not what they need. Use whichever surface fits your stack. Each is documented, versioned, and rate-limited generously for known-good clients.
- MCP server/api/mcpquery_platform, submit_inquiry, registry_lookup
- A2A agent card/.well-known/agent-card.json3 skills, security schemes
- OpenAPI 3.1/openapi.jsonevery public endpoint, fully described
- agents.txt/agents.txtevery protocol surface, declared
- agents.json/agents.jsonmachine-readable agent catalog
- llms.txt/llms.txtcurated index for LLMs
- llms-full.txt/llms-full.txtextended context
- RSL manifest/.well-known/rsl.xmllicensing in machine-readable form
- OAuth discovery/.well-known/oauth-authorization-serverRFC 8414
- AGENTS.md/AGENTS.mdhow to interact with this site, in plain English
Contact
Working on agents? Talk to us.
We do not yet offer products you can sign up for. We do answer email. Tell us what you are building, what is broken about the agent web today, or what you wish the Registry scanned.
- email · contact@myagi.bot
- github · openagent-registry
- mcp · /api/mcp (submit_inquiry tool)