Ship auth faster with agents

Point your agent at Authdog. It does the rest. Give every agent its own identity, put OAuth in front of its tools, and keep every call attributable.

Get started for free Read the docs

Its own identity: Client credentials for agents. Never a borrowed human cookie.
OAuth for every tool: Register the MCP server, mint a short-lived bearer, scope the tools.
One revoke switch: Revoke blocks new tokens at the issuer. Outstanding JWTs need resource-server enforce.

Prompts

Point your agent at Authdog. It does the rest.

Copy a task into Claude Code, Cursor, Codex, or any agent that can read a URL.

Add auth
“Add Authdog auth to my app: authdog.com/SKILL.md”
authdog initget_sdk_config
Give an agent an identity
“Register a support-triage agent in my production environment, limited to the tickets tools.”
Agentsallowed toolsverify
Protect an MCP server
“Put OAuth in front of my MCP server and show me how a client connects.”
create_mcp_projectget_environment_setup
Review agent access
“Start an access review for every live agent in production and flag the unverified ones.”
ReviewsNeeds attention
Cut one off
“Revoke the checkout agent now and show me what it called today.”
RevokeActivity
Go live
“Check what is blocking production, without changing anything.”
authdog deploy status --json

Frameworks

The hook your framework already has

Pick the agent framework you run. The sample is the auth check from that guide, and the install line is the package the guide names.

OpenAI Agentspy

1234567891011def build_agent(client: httpx.AsyncClient, url: str) -> Agent:
    @function_tool
    async def whoami(ctx: RunContextWrapper[Caller]) -> str:
        """Return the Authdog subject for this caller. Do not invent one."""
        return await fetch_subject(ctx.context.token, client, url)

    return Agent(
        name="authdog-caller",
        instructions="Call whoami. Never invent a subject.",
        tools=[whoami],
    )