# Postcept

> Postcept is an MCP server for agent approvals. Verifies agent actions like refunds and cancellations against your systems of record. It runs locally on your machine and is started by your AI client.

- **Kind:** MCP server
- **Platforms:** API, CLI, Self-hosted
- **Repository:** https://github.com/postcept/mcp
- **MCP Registry name:** io.github.Postcept/mcp
- **Categories:** [agent approvals](https://ai-powered.com/categories/agent-approvals)
- **Listing updated:** 2026-11-21

## About

Postcept is an MCP server that lets an AI agent check whether a high-risk action, such as a refund, a subscription cancellation or a support ticket resolution, really completed in the underlying system. Instead of trusting the agent's own report that work is done, the verification tools query the system of record, including Stripe and Zendesk, and return a signed completion receipt.

The server is a thin wrapper over the Postcept HTTP API and needs only a Postcept API key with verification write access. Verification is read-only against connected systems, and a sandbox connector lets developers test the full flow without touching live data.

It suits teams building support or billing agents in development, demos, or workflows where the agent's judgment is already trusted. The documentation notes that an agent can skip the call or ignore the result, so for consequential actions the same API can be called from an orchestrator that gates customer-facing steps.

## Features

- **Refund verification:** Checks that a claimed refund exists with matching amount, currency and customer, and is not duplicated.
- **Cancellation verification:** Confirms that a subscription was actually cancelled in the system of record.
- **Ticket state verification:** Confirms a support ticket is in the expected state after an agent resolves it.
- **Signed completion receipts:** Returns a signed receipt for each verification that can be fetched later by id.
- **Re-verification:** Re-checks a past verification against the live system to detect changes.
- **Sandbox testing mode:** Runs verifications against a deterministic sandbox connector excluded from completion metrics.

## Good for

- Confirming an agent-issued refund was actually processed before telling a customer
- Verifying subscription cancellations completed in billing before closing a request
- Checking that support tickets an agent resolved are in the expected state
- Exercising an agent's verification flow in development without live systems

## Install

### Claude Code (Terminal)

```sh
claude mcp add mcp --env POSTCEPT_API_KEY='<POSTCEPT_API_KEY>' -- npx -y @postcept/mcp
```

### Claude Desktop (claude_desktop_config.json)

```json
{
  "mcpServers": {
    "mcp": {
      "command": "npx",
      "args": [
        "-y",
        "@postcept/mcp"
      ],
      "env": {
        "POSTCEPT_API_KEY": "<POSTCEPT_API_KEY>"
      }
    }
  }
}
```

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

```json
{
  "mcpServers": {
    "mcp": {
      "command": "npx",
      "args": [
        "-y",
        "@postcept/mcp"
      ],
      "env": {
        "POSTCEPT_API_KEY": "<POSTCEPT_API_KEY>"
      }
    }
  }
}
```

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

```json
{
  "servers": {
    "mcp": {
      "type": "stdio",
      "command": "npx",
      "args": [
        "-y",
        "@postcept/mcp"
      ],
      "env": {
        "POSTCEPT_API_KEY": "<POSTCEPT_API_KEY>"
      }
    }
  }
}
```

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

```json
{
  "mcpServers": {
    "mcp": {
      "command": "npx",
      "args": [
        "-y",
        "@postcept/mcp"
      ],
      "env": {
        "POSTCEPT_API_KEY": "<POSTCEPT_API_KEY>"
      }
    }
  }
}
```

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

```json
{
  "mcpServers": {
    "mcp": {
      "command": "npx",
      "args": [
        "-y",
        "@postcept/mcp"
      ],
      "env": {
        "POSTCEPT_API_KEY": "<POSTCEPT_API_KEY>"
      }
    }
  }
}
```

## Tools

- `verify_refund`: Verify a claimed refund (exists, amount/currency/customer match, not duplicated).
- `verify_cancellation`: Verify a subscription was actually cancelled.
- `verify_ticket`: Verify a support ticket is in the expected state.
- `get_verification`: Fetch a past verification and its signed receipt by id.
- `reconcile_verification`: Re-verify a past verification against the live system of record.
- `verified_completion_rate`: Read the organization's Verified Completion Rate.

## FAQ

### What do I need to run the server?

A Postcept API key with verifications:write permission, set as the POSTCEPT_API_KEY environment variable. It is launched over stdio with npx.

### Can I point Postcept at a self-hosted API?

Yes, the POSTCEPT_BASE_URL variable can be set to a self-hosted or staging API endpoint.

### What does a non-verified result mean?

Results such as incomplete, duplicated, mismatched or policy_failed mean the work is not done, and the agent should surface the gap and recover.

### How do I add Postcept to Claude Code?

Run this command in your terminal: claude mcp add mcp --env POSTCEPT_API_KEY='<POSTCEPT_API_KEY>' -- npx -y @postcept/mcp

### What are the best alternatives to Postcept?

The closest alternatives to Postcept are Attest and Proctor.

## Where this comes from

- Description: written from the project's README and its MCP Registry entry.
- Install steps: from the official MCP Registry entry.

## Alternatives

- [Attest](https://ai-powered.com/mcp/attest-2): Attest is an open-source MCP server for agent approvals.
- [Proctor](https://ai-powered.com/mcp/proctor-2): Proctor is an open-source MCP server for security.

---
Source: https://ai-powered.com/mcp/postcept
