Proofable MCP

Hosted MCP server for AI agent identity, permissions, verification, and reusable proof. Connect an AI client, sign in with OAuth, and check what an agent may do before it acts.

Connect

Add the hosted server to your MCP client, then click Connect and sign in:

{
  "mcpServers": {
    "proofable": { "url": "https://mcp.proofable.me/mcp" }
  }
}

Or let the installer write the same entry for Cursor, Claude Code, Codex, or VS Code:

npx -y @proofable/sdk setup

Then ask: "Show my Proofable profile and current proofs."

This repository is also a plugin marketplace. In Claude Code:

/plugin marketplace add proofable/mcp
/plugin install proofable-mcp@proofable

The plugin registers the server and ships the skills in plugins/proofable-mcp. Use either the plugin or a manual entry, not both.

What it does

Job	Tools
Load the signed-in profile and workflow	proofable_context (call first)
Check, reuse, or create proof	proofable_proofs_check, proofable_verify_or_guide, proofable_verify, proofable_proofs_get, proofable_verifiers_catalog
Give agents an owner and permissions	proofable_agent_link, proofable_agent_create, proofable_agent_mount
Store secrets without exposing them	proofable_secret_create, proofable_secret_list, proofable_secret_revoke

Full reference: docs.proofable.me/mcp/tools.

Authentication

Interactive clients sign in with OAuth. Servers and CI use a profile access key in PROOFABLE_ACCESS_KEY. Never put a key in client config or chat. See OAuth.

This package

@proofable/mcp publishes the registry manifest (server.json) and the public skills. It does not run a local server.

import { serverManifest } from '@proofable/mcp';

Support

• Docs: docs.proofable.me/mcp/overview
• Issues: github.com/proofable/mcp/issues
• Security: SECURITY.md

Apache-2.0. Proofable is published by NEUS Network, Inc.