# AgentAdmit

> AgentAdmit is user-mediated authorization for AI agents: a hosted service that lets an application grant an AI agent scoped, revocable, auditable access to a user's data, with the human user in the loop for every grant. It complements standards like OAuth. Patent pending.

AgentAdmit is live in production in TrainerTracer (web and both mobile app stores). Server SDKs are published for Node.js, Python, Go, Ruby, PHP, and Java, plus a React component library.

## Start here

- [Home](https://agentadmit.com/): what AgentAdmit is and how the flow works
- [Documentation](https://agentadmit.com/docs): integration guides and reference
- [App Owner Guide](https://agentadmit.com/docs/app-owner-guide): register an app, define scopes, issue and verify connections, revoke, protections, consent ledger
- [MCP Server Guide](https://agentadmit.com/docs/mcp-guide): give an MCP server or agent tool safe, user-approved access to a user's account
- [For coding agents](https://agentadmit.com/coding-agents): test mode, limits, and how coding agents integrate
- [Pricing](https://agentadmit.com/pricing): per-active-connection plans; test mode for development
- [FAQ](https://agentadmit.com/faq)

## Terms we define

- [Glossary](https://agentadmit.com/glossary): canonical definitions of the terms below
- User-Mediated Authorization: the authorization credential is delivered to the human user through a channel the user controls, and the user hands it to their agent out of band. There is no redirect, callback, or webhook an agent can reach, so a prompt-injected agent cannot obtain or escalate access on its own.
- Self-Describing Credential: a connection credential that encodes where it should be exchanged, so an agent can connect to an application it has never seen before.
- Discovery-by-Introspection: exchanging or introspecting a credential returns scope-filtered operational metadata, so an agent sees only the endpoints it is authorized to call.
- Caller-Identity Consent: one API surface serves several classes of caller (the human, the app's in-app AI, an external agent) and the data owner consents to each class independently.
- Verifiable Consent Evidence: a consent record whose proof does not depend on trusting anyone's database; the passkey signature covers a cryptographic commitment to the recorded consent parameters.

## Compliance and policies

- [SDK Privacy and Compliance Guide](https://agentadmit.com/docs/compliance): app store disclosure answers and data handling
- [Privacy Policy](https://agentadmit.com/docs/privacy)
- [Terms of Service](https://agentadmit.com/docs/terms)

## Blog

- [AgentAdmit Blog](https://blog.agentadmit.com/): articles on AI agent authorization, MCP security, and prompt injection as an authorization problem. See https://blog.agentadmit.com/llms.txt for the article index.

## Contact

- Email: christopher@agentadmit.com
- X: https://x.com/AgentAdmitHQ