Endram field guides

AI agent policy templates

Copyable starting policies for production actions, with rollout and evidence guidance.

33 implementation pages5 sectionsUpdated September 2026

MCP protocol reference

01 · rfc 9728

MCP protected resource metadata (RFC 9728)

The exact protected resource metadata document, WWW-Authenticate challenge, and discovery path an MCP server needs so a client can find its authorization server without configuration.

Open guide →
02 · mcp dynamic client registration

MCP dynamic client registration without a shared secret

How an MCP server accepts RFC 7591 registration from clients it has never met, which fields to reject, and why the only safe answer for a desktop agent is a public PKCE client.

Open guide →
03 · rfc 8707

Binding MCP tokens to one server with resource indicators (RFC 8707)

How the resource parameter stops a token minted for one MCP server from working against another, with the exact authorization and token requests and the server-side check that makes it real.

Open guide →
04 · mcp token passthrough

MCP token passthrough and the confused deputy

Why forwarding a client's token to an upstream API is explicitly disallowed by the MCP specification, what breaks when you do it, and the exchange pattern that replaces it.

Open guide →
05 · mcp streamable http

Securing MCP Streamable HTTP transport

The transport-level checks a Streamable HTTP MCP server needs: Origin validation, local binding, session handling, and what changed when SSE became a response mode instead of a transport.

Open guide →
06 · mcp session id

MCP session ID: what it is for and what it must never be

How the Mcp-Session-Id header is issued, carried, and terminated, why it is not an authentication credential, and the failures that follow from treating it as one.

Open guide →
07 · mcp elicitation

MCP elicitation: what to allow a server to ask a user for

Elicitation lets an MCP server request input from the human mid-call. The controls that matter are what may be asked, what the answer is allowed to change, and why an elicited answer is not an approval.

Open guide →
08 · mcp sampling

MCP sampling: letting a server borrow your model safely

Sampling lets an MCP server ask the client to run a model completion. The controls are cost, prompt provenance, and making sure a server-authored completion never becomes an authorization decision.

Open guide →
09 · oauth token exchange

OAuth token exchange for agent delegation (RFC 8693)

The request and response shape for exchanging a user's token for a narrow downstream token an agent can use, and the claims that keep the human visible on the other side.

Open guide →

Implementation and testing

01 · mcp server

MCP server security checklist

A production-ready starting policy for mcp server, including match conditions, approval behavior, evidence, and rollout checks.

Open guide →
02 · mcp server examples

MCP server examples with security policy

A production-ready starting policy for mcp server examples, including match conditions, approval behavior, evidence, and rollout checks.

Open guide →
03 · mcp server authentication

MCP server authentication checklist

A pass or fail checklist for authenticating a remote MCP server: discovery, registration, PKCE, audience binding, scope enforcement, revocation, and the tests that prove each one.

Open guide →
04 · mcp tool poisoning

MCP tool poisoning and description drift

Tool descriptions are model-visible instructions. This is how a server changes them after approval, why the client cannot see it, and the pinning and diffing controls that catch it.

Open guide →
05 · fastmcp auth

FastMCP authentication for a remote server

What a FastMCP server has to expose before an external client can authenticate to it, which parts FastMCP gives you, and which parts remain your authorization server's job.

Open guide →
06 · mcp remote

Connecting a client to a remote MCP server with mcp-remote

The bridge that lets a stdio-only MCP client reach an OAuth-protected remote server, the config that works, and the failures that come from stale credentials rather than from your server.

Open guide →
07 · mcp inspector

Testing an authenticated MCP server with MCP Inspector

How to use Inspector to prove that discovery, the OAuth flow, scopes, and audience binding actually work on your MCP server, instead of testing only the tools.

Open guide →
08 · langgraph human in the loop

LangGraph human-in-the-loop approval that survives a restart

How interrupt and Command resume a LangGraph run for human review, why the checkpointer is the part that matters, and where the approval decision should actually live.

Open guide →

MCP policy templates

Action policy templates

01 · github ai agent approval

GitHub merge approval policy

A production-ready starting policy for github ai agent approval, including match conditions, approval behavior, evidence, and rollout checks.

Open guide →
02 · ai agent deployment approval

Production deployment approval policy

A production-ready starting policy for ai agent deployment approval, including match conditions, approval behavior, evidence, and rollout checks.

Open guide →
03 · ai agent refund approval

Agent refund threshold policy

A production-ready starting policy for ai agent refund approval, including match conditions, approval behavior, evidence, and rollout checks.

Open guide →
04 · ai agent database access policy

AI agent database write policy

A production-ready starting policy for ai agent database access policy, including match conditions, approval behavior, evidence, and rollout checks.

Open guide →
05 · ai agent email approval

AI agent external email policy

A production-ready starting policy for ai agent email approval, including match conditions, approval behavior, evidence, and rollout checks.

Open guide →
06 · ai agent slack policy

AI agent Slack announcement policy

A production-ready starting policy for ai agent slack policy, including match conditions, approval behavior, evidence, and rollout checks.

Open guide →
07 · ai agent kill switch

AI agent emergency kill switch policy

A production-ready starting policy for ai agent kill switch, including match conditions, approval behavior, evidence, and rollout checks.

Open guide →

Delegation and credentials

01 · verified ai agent identity policy

Verified AI agent identity policy

A production-ready starting policy for verified ai agent identity policy, including match conditions, approval behavior, evidence, and rollout checks.

Open guide →
02 · ai agent delegation policy

Delegated AI agent task policy

A production-ready starting policy for ai agent delegation policy, including match conditions, approval behavior, evidence, and rollout checks.

Open guide →
03 · task scoped credentials github

Task-scoped GitHub credential policy

A production-ready starting policy for task scoped credentials github, including match conditions, approval behavior, evidence, and rollout checks.

Open guide →
04 · one time access token ai agent

One-use AI agent grant policy

A production-ready starting policy for one time access token ai agent, including match conditions, approval behavior, evidence, and rollout checks.

Open guide →
05 · purpose bound access policy

Purpose-bound AI agent access policy

A production-ready starting policy for purpose bound access policy, including match conditions, approval behavior, evidence, and rollout checks.

Open guide →
06 · expiring agent delegation

Expiring AI agent delegation policy

A production-ready starting policy for expiring agent delegation, including match conditions, approval behavior, evidence, and rollout checks.

Open guide →
07 · ai agent least privilege policy

AI agent least-privilege review policy

A production-ready starting policy for ai agent least privilege policy, including match conditions, approval behavior, evidence, and rollout checks.

Open guide →