agentfabric.dev
MCP server live AI agents connect in one line →

Human-in-the-loop for AI agents

Send agent output to humans. Get approval, feedback, and an audit trail.

agentfabric.dev is the review and approval layer for AI agents. Route generated URLs, images, and documents to human reviewers, enforce approval policies, and keep a complete record of who signed off — over MCP or REST.

Connect your agent Read the docs 
Loading…

For AI agents

Built for agents first

agentfabric.dev exposes a native Model Context Protocol server, so any MCP-compatible agent gets human review as a set of tools — no glue code required.

MCP endpoint https://mcp.agentfabric.dev

Add this server to Cursor, Claude, or any MCP client. The agent can then create a tenant, open review requests, poll for decisions, and read reviewer comments.

create_tenant create_review_request get_review_request list_review_requests list_comments add_reviewer
  1. 1

    Add the MCP server

    Drop the endpoint into your MCP client config. No SDK, no build step.

  2. 2

    Call create_tenant

    The agent gets admin and consumer API keys back automatically on first run.

  3. 3

    Request human review

    Send content with create_review_request, then poll get_review_request for the verdict.

MCP-native onboarding

Agents connect over the Model Context Protocol and get review tools instantly. Predictable JSON, no surprises.

Transparent audit trail

Every comment, approval, and rejection is recorded per reviewer — accountable, traceable decisions for autonomous systems.

Policies you control

All must approve, any approval, quorum, or first rejection wins — at the request level or per item.

FAQ

Questions agents and builders ask

What is agentfabric.dev?

A human-in-the-loop review and approval layer for AI agents. Agents send generated content — URLs, images, PDFs, or documents — to human reviewers who approve, reject, or comment. Every decision lands in an audit trail.

How does an AI agent connect?

The fastest path is MCP. Point any MCP-compatible client at the agentfabric.dev MCP server and the agent gets tools to create a tenant, open review requests, poll results, and read comments. A REST API is available for backend services.

Do human reviewers need an account?

No. Reviewers get an email with a personal, scoped link. They comment, approve, or reject in the browser — no signup.

What approval policies are supported?

ALL_APPROVE, ANY_APPROVE, ANY_REJECT, and QUORUM. Decisions can be scoped to the whole request or to individual items.

Ready to ship with oversight?

Landing pages, creatives, documents — ship with human oversight and a clear record of who signed off.

Connect your agent Sign up