# Install the AgentProof MCP server

Connect AgentProof to Claude, ChatGPT, Cursor, Codex, VS Code or Slackbot. Sign in once, then start audits and read findings from chat.

The AgentProof MCP server lets an AI assistant work with your AgentProof account. You can start a test of your live website, check a run, read reports and findings, and manage your workspace and its members from the chat you already use.

Every assistant uses the same address:

```
https://isagentproof.com/mcp
```

The first time you connect, the assistant opens AgentProof in your browser. Sign in, or create an account with your email. Then press **Allow** to give the assistant access. You never paste a password or an API key into the chat.

## Claude.ai and Claude Desktop

1. Open **Settings**, then **Connectors**.
2. Choose **Add custom connector**.
3. Enter the name AgentProof and the address `https://isagentproof.com/mcp`.
4. Press **Connect**, sign in to AgentProof and press **Allow**.

On Team and Enterprise plans, an owner adds the connector for the organization.

## Claude Code

Run this command in your terminal:

```
claude mcp add --transport http agentproof https://isagentproof.com/mcp
```

Then type `/mcp` in Claude Code and choose **agentproof** to sign in.

You can also install the AgentProof plugin, which adds the MCP server and the remote audit skills together:

```
/plugin marketplace add GravyPouch/agentproof-skills
/plugin install agentproof-remote-audits@agentproof-skills
```

## ChatGPT

1. Open **Plugins** in ChatGPT.
2. Choose **+**, then **Add custom MCP server**.
3. Enter the address `https://isagentproof.com/mcp`.
4. Sign in to AgentProof and press **Allow**.

## Cursor

Add AgentProof to `.cursor/mcp.json` in your project, or to `~/.cursor/mcp.json` for every project:

```json
{
  "mcpServers": {
    "agentproof": {
      "url": "https://isagentproof.com/mcp"
    }
  }
}
```

Cursor asks you to sign in the first time it uses a tool.

## VS Code

Add AgentProof to `.vscode/mcp.json`:

```json
{
  "servers": {
    "agentproof": {
      "type": "http",
      "url": "https://isagentproof.com/mcp"
    }
  }
}
```

## Codex

Add AgentProof to `~/.codex/config.toml`:

```toml
[mcp_servers.agentproof]
url = "https://isagentproof.com/mcp"
```

## Other MCP clients

Add a remote server that uses Streamable HTTP, with the address `https://isagentproof.com/mcp`. AgentProof signs in with OAuth 2.1. Clients register themselves, so you do not need a client ID or secret.

## Slackbot

Slackbot can call AgentProof's tools when your workspace has its own AgentProof Slack app.

1. Go to [api.slack.com/apps](https://api.slack.com/apps) and choose **Create New App**, then **From a manifest**.
2. Pick your Slack workspace and paste this manifest:

```json
{
  "display_information": {
    "name": "AgentProof",
    "description": "Run AgentProof audits and read reports and findings from Slackbot."
  },
  "features": {
    "bot_user": { "display_name": "AgentProof", "always_online": true }
  },
  "oauth_config": {
    "scopes": { "bot": ["mcp:connect"] }
  },
  "settings": {
    "org_deploy_enabled": true,
    "socket_mode_enabled": false,
    "token_rotation_enabled": false
  },
  "mcp_servers": {
    "agentproof": {
      "url": "https://isagentproof.com/mcp",
      "auth_provider_key": "mcp_dcr_agentproof",
      "auth_type": "dynamic_client_registration"
    }
  }
}
```

3. Install the app to your workspace.
4. Ask Slackbot about AgentProof. The first time, it asks you to sign in to AgentProof and press **Allow**.

Slackbot asks you to approve each tool call, and it stops waiting after 60 seconds. Audits start right away and run in the background, so ask Slackbot for the run or its findings later.

## What the assistant can do

The access screen lists what you allow. These are all of the tools:

| Tool | What it does |
| --- | --- |
| `get_account` | Shows your account and your workspaces. |
| `start_audit` | Starts an Attack Agent test of a website you own or may test. Uses one credit. |
| `start_run` | Starts a task on one of your agents. Uses one credit. |
| `get_run` | Shows a run's status and its page. |
| `list_reports` | Lists finished runs with reports. |
| `list_findings` | Lists findings with their severity and status, and the fix pull request. |
| `list_agents` | Lists your Attack and Defense agents. |
| `get_billing` | Shows your plan status and credits left. |
| `list_members` | Lists workspace members and pending invitations. |
| `invite_member` | Emails an invitation to your workspace. |
| `change_member_role` | Makes a member an admin or a member. |
| `remove_member` | Removes someone from your workspace. |
| `revoke_invitation` | Cancels a pending invitation. |
| `leave_workspace` | Removes you from a workspace you do not own. |
| `create_workspace` | Creates a workspace that you own. |
| `rename_workspace` | Renames a workspace you own. |
| `list_notification_destinations` | Lists where your workspace sends notifications. |
| `send_test_notification` | Sends a test message to one destination. |

Paying, changing your plan, transferring ownership and deleting a workspace or account happen only on this website. Tools return names, statuses, links and finding titles. Agent replies and evidence stay on the run page.

## Disconnect an assistant

Open **Settings**, then **Security**. Under **Connected apps**, press **Disconnect** next to the assistant. It loses access at once.