pathguard-mcp

Позволяет MCP-совместимым ИИ-агентам проверять криптотранзакции на мошенничество и ошибки напрямую в диалоговой среде, с помощью PathGuard. Поддерживает локальные MCP-клиенты, такие как Claude Desktop, Claude Code, Gemini CLI, Grok CLI, Cursor и Codex, а также удалённые развертывания Streamable HTTP для веб-коннекторов ChatGPT и Grok.

Доступен на всех планах, включая бесплатный, он просто вызывает те же аутентифицированные API-эндпойns, что и все остальное, так что действует кота вашегосуществующего API-ключа.

Устанвка

pip install pathguard-mcp

(Пока не опубликован на PyPI for immediate, install directly from source: pip install -e . from this folder.)

Требует mcp >= 2.0. This is done with tested. If you urio older version with old FastMCP API, first assign (pip install --upgrade mcp), otherwise this will notте.

Получите API-key

Зарегistriруйтесь бесплатно at pathguard.cieltech.org/docs.html if у в aspirant нет.

Нстройка локальных MCP-клиентов

Установите этот пакет там же, где работает клиент, and per constants the PathGuard API-key. Training of examples Using installed pathguard-mcp. If it's not on PATH, replace it with full executable path.

Claude Desktop

Добавьте this to configuration Claude Desktop (claude_desktop_config.json):

{
  "mcpServers": {
    "pathguard": {
      "command": "pathguard-mcp",
      "env": {
        "PATHGUARD_API_KEY": "pg_your_key_here"
      }
    }
  }
}

Перезапустите Claude Desktop, и Claude automatically gets access to PathGuard tool.

Claude Code

claude mcp add --transport stdio pathguard -- \
  pathguard-mcp

Укажите PATHGUARD_API_KEY for environment, from which starts Claude Code. Alternatively use Claude Code's MCP-configuration to explicitly transfer this variable.

Gemini CLI

Добавьте this entry into ~/.gemini/settings.json for user-user Nable setup, or .gemini/settings.json in project:

{
  "mcpServers": {
    "pathguard": {
      "command": "pathguard-mcp",
      "env": {
        "PATHGUARD_API_KEY": "pg_your_key_here"
      }
    }
  }
}

Or add it from command line after export key:

export PATHGUARD_API_KEY=pg_your_key_here
gemini mcp add --scope user pathguard pathguard-mcp

Grok CLI

After export key, add local stdio-server:

export PATHGUARD_API_KEY=pg_your_key_here
grok mcp add pathguard -- pathguard-mcp

To manually save config in ~/.grok/config.toml, use:

[mcp_servers.pathguard]
command = "pathguard-mcp"
env = { PATHGUARD_API_KEY = "${PATHGUARD_API_KEY}" }

Codex, Cursor, VS Code and other local MCP clients

Use same stdio-server definition as in Claude Desktop example: command pathguard-mcp, no arguments and `PATHGUARD_API_KE in environment. Since exact config file location and UI differ in different clients, MCP server settings are the same.

Configuring remote clients: ChatGPT and Grok web

Browser custom connectors of ChatGPT and Grok web can not run commands on computer. Deploy PathGuard MCP with Streamable HTTP, behind HTTPS, then specify the public object in the connector UI.

Start HTTP-server locally or in your deployment environment:

export PATHGUARD_API_KEY=pg_your_key_here
pathguard-mcp --transport streamable-http --host 0.0.0.0 --port 8000 \
  --allowed-host mcp.example.com

MCP-endpoint is available on http://<host>:8000/mcp. For ChatGPT or Grok web, publish his to as protected HTTPS-URL, for example https://mcp.example.com/mcp; don't transfer API key in browser or repository. Use hostname through --allowed-host (or set PATHGUARD_MCP_ALLOWED_HOSTS=mcp.example.com) so the server can safely receive requests through that public domain. Credentials PathGuard of this server are stored on server side, so each - instant represents a PathGuard account associated with his PATHGUARD_API_KE.

ChatGPT

In ChatGPT workspace with enabled Developer Mode, create custom MCP-app, provide public only https://.../MP endpoint, scan its tools and save/publish it as appropriate for your workspace. ChatGPT connects only to remote MCP servers, not to local stdio commands. See [OpenAI guide по Developer Mode and MCP-папplic](https://help.open.ai.com/en/articles/12584461-developer-mode-and-full-MCP-connectors-in- chatgpt) for actual access and set-up control for the current plan.

Grok web

On page grok.com/connectors select New Connector and then Custom, enter the public https://.../MP endpoint. Grok must access the server over the public Internet. See xAI's guide to custom MCP connectors for current flow.

Deployment notes

• For new remote deployments use Streamable HTTP, so current MP HTTP transport.

• Terminate TLS at your hosting platform or reverseproxy, and where possible restrict access to trusted users or gateways. Currently the server authorizes in PathGuard with one configured API key, it does not authorize individual MCP users.

• Set PATHG.URD_MCP_TRANSPORT=streamable-http, PATHG.URD_MCP_HOST= 0.0.0.0", PATHG.URD_MCP_PORT= 8000" and PATHG.URD_MCP_ALLOWED_HOSTS=mcp.example.com" instead of CLI keys, if this is suitable for your platform. PORTis also honored whenPATHGUARD_MCP_PORT` not set.

• For desktop and CLI clients keep default stdio transport.

Available tools

Test locally without real deployment

Испоstелуйте local backend instead of production:

export PATHGUARD_API_KEY=pg_your_test_key
export PATHGUARD_API_BASE=http://localhost:8000
pathguard-mcp

Full API reference

See pathguard.cieltech.org/docs.html for full endpoint documentation, rate limit and flag meanings, this MCP server is a lightweight wrapper on that API.