# agent-custody

> agent-custody is a freemium MCP server for security. Gateway that checks agent tool calls against signed grants and records signed receipts. It runs locally on your machine and is started by your AI client.

- **Kind:** MCP server
- **Pricing:** Freemium
- **Starting price:** $50 a month
- **Platforms:** CLI, self-hosted, API
- **Website:** https://agent-custody.dev
- **Repository:** https://github.com/svayatta/agent-custody/tree/HEAD/packages/receipts
- **MCP Registry name:** io.github.svayatta/agent-custody
- **Categories:** security, developer tools
- **Listing updated:** 2026-11-14

## About

The idea is a tamper-evident trail for what an AI agent actually did. A hook sits between the agent and its tools, checks each shell command, file edit or MCP call against a policy written as deny rules, and writes a signed receipt whose hash goes into an append-only log. Anyone holding the public keys can verify a receipt offline.

One command sets up a key, a starter deny list and Claude Code hooks for the current directory. It begins in observe mode, so nothing is blocked until you remove one line from the config. An audit-sessions command reads your past Claude Code transcripts and reports which calls the starter list would have caught, writing nothing. The site's own example shows force pushes and terraform destroy being flagged.

The gateway, SDK, ledger and verifier are Apache-2.0 and free to run yourself. What is priced is a hosted log run by a third party, so outsiders can trust the record. It needs Node 22.

## Features

- **Signed receipts:** Each call produces a receipt that can be verified offline with public keys.
- **Deny-list policy:** Rules written as policies block calls such as force pushes or terraform destroy.
- **Observe mode first:** Starts by recording only, so you can see what would be denied.
- **Transcript audit:** Replays past Claude Code sessions against the policy without writing anything.
- **Tamper-evident log:** Receipt hashes land in a log that others can check.
- **Hosted log option:** A paid, third-party-run log lets outsiders accept your records.

## Good for

- Prove to a reviewer which commands an agent ran
- Block destructive commands like force pushes before they execute
- Find out what a deny list would have caught in last week's sessions
- Give auditors verifiable records of agent activity

## Install

### Claude Code (Terminal)

```sh
claude mcp add agent-custody -- npx -y @agent-custody/receipts
```

### Claude Desktop (claude_desktop_config.json)

```json
{
  "mcpServers": {
    "agent-custody": {
      "command": "npx",
      "args": [
        "-y",
        "@agent-custody/receipts"
      ]
    }
  }
}
```

### Cursor (~/.cursor/mcp.json)

```json
{
  "mcpServers": {
    "agent-custody": {
      "command": "npx",
      "args": [
        "-y",
        "@agent-custody/receipts"
      ]
    }
  }
}
```

### VS Code (.vscode/mcp.json)

```json
{
  "servers": {
    "agent-custody": {
      "type": "stdio",
      "command": "npx",
      "args": [
        "-y",
        "@agent-custody/receipts"
      ]
    }
  }
}
```

### Windsurf (~/.codeium/windsurf/mcp_config.json)

```json
{
  "mcpServers": {
    "agent-custody": {
      "command": "npx",
      "args": [
        "-y",
        "@agent-custody/receipts"
      ]
    }
  }
}
```

### Other clients (your client's MCP settings (most use this shape))

```json
{
  "mcpServers": {
    "agent-custody": {
      "command": "npx",
      "args": [
        "-y",
        "@agent-custody/receipts"
      ]
    }
  }
}
```

## FAQ

### Does it block anything straight away?

No. It starts in observe mode and blocks only after you remove the observe setting.

### Can I verify receipts without the vendor?

Yes. Anyone with the public keys can verify a receipt offline.

### What does it cost?

The self-run parts are Apache-2.0 and free. The hosted log has a Free tier and a Team plan at $50 a month.

### What runtime does it need?

Node 22.

### Is agent-custody free?

agent-custody has a free plan, with paid plans for more usage or features. Paid plans start at $50 a month.

### How do I add agent-custody to Claude Code?

Run this command in your terminal: claude mcp add agent-custody -- npx -y @agent-custody/receipts

### What are the best alternatives to agent-custody?

The closest alternatives to agent-custody are REA, Codebase Memory, GitHub, Serena MCP: the IDE for your agent and XERJ.

## Alternatives

- [REA](https://ai-powered.com/mcp/rea): REA is an open-source MCP server for developer tools.
- [Codebase Memory](https://ai-powered.com/mcp/codebase-memory): Codebase Memory is an open-source MCP server for knowledge and memory.
- [GitHub](https://ai-powered.com/mcp/github-2): GitHub is an open-source MCP server for developer tools.
- [Serena MCP: the IDE for your agent](https://ai-powered.com/mcp/serena-mcp-the-ide-for-your-agent): Serena MCP: the IDE for your agent is an open-source MCP server for developer tools.
- [XERJ](https://ai-powered.com/mcp/xerj): XERJ is an open-source MCP server for knowledge and memory.
- [Ouroboros](https://ai-powered.com/mcp/ouroboros): Ouroboros is an open-source MCP server for developer tools.
- [Claude Code Ultimate Guide](https://ai-powered.com/mcp/claude-code-ultimate-guide): Claude Code Ultimate Guide is an open-source MCP server for ai models and llm tooling.
- [Concord MCP](https://ai-powered.com/mcp/concord-mcp): Concord MCP is a freemium MCP server for developer tools.

---
Source: https://ai-powered.com/mcp/agent-custody
