# Aggrete MCP server

> Enforces a code-of-conduct policy across MCP connectors, refusing forbidden tool combinations before the upstream call runs.

- Repository: https://github.com/Aggrete/aggrete
- Page: https://hostedmcp.org/servers/aggrete/aggrete
- Transport: stdio
- Stars: 2
- License: Apache-2.0
- Latest version: 0.8.3

## Install

### Claude Code

```sh
claude mcp add aggrete -- uvx aggrete
```

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

```json
{
  "mcpServers": {
    "aggrete": {
      "command": "uvx",
      "args": [
        "aggrete"
      ]
    }
  }
}
```

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

```json
{
  "servers": {
    "aggrete": {
      "type": "stdio",
      "command": "uvx",
      "args": [
        "aggrete"
      ]
    }
  }
}
```

### VS Code CLI

```sh
code --add-mcp '{"name":"aggrete","type":"stdio","command":"uvx","args":["aggrete"]}'
```

### Claude Desktop (`claude_desktop_config.json`)

```json
{
  "mcpServers": {
    "aggrete": {
      "command": "uvx",
      "args": [
        "aggrete"
      ]
    }
  }
}
```

### Codex

```sh
codex mcp add aggrete -- uvx aggrete
```

### Codex config.toml (`~/.codex/config.toml`)

```toml
[mcp_servers.aggrete]
command = "uvx"
args = ["aggrete"]
```

### Gemini CLI

```sh
gemini mcp add -s user aggrete uvx -- aggrete
```

### Gemini settings.json (`~/.gemini/settings.json`)

```json
{
  "mcpServers": {
    "aggrete": {
      "command": "uvx",
      "args": [
        "aggrete"
      ]
    }
  }
}
```

## FAQ

### What does Aggrete do for MCP clients?

It acts as a proxy that checks each tool call against a code-of-conduct policy and per-user memory, refusing forbidden calls before the data is fetched.

### How do I dry-run a plan of tool calls?

Use the check tool, which returns an allowed or REFUSED verdict for a sequence of calls before anything is fetched.

### Can Aggrete detect issues in my existing MCP setup?

Yes, run aggrete scan to find cross-server leak paths, hidden instructions in descriptions, secrets in client configs, and proxy bypasses.

### Does Aggrete redact sensitive data from results?

It redacts emails, SSNs, cards, and tokens from tool results before they reach the model.

### How are decisions audited?

Every decision is written as a hash-chained JSON audit line and can be forwarded to a SIEM or OpenTelemetry collector.
