Model Context Protocol (MCP)

An MCP server is a lightweight program that implements the Model Context Protocol (MCP), enabling AI agents to securely interact with external systems — APIs, databases, or infrastructure tools.

ZenSailor exposes a secure MCP endpoint (/api/mcp, Streamable HTTP with Bearer auth) that lets you control your PaaS from tools like Cursor, Windsurf, Claude Desktop or Codex using natural language.

What the tools can do#

• List services — “What services are running?” or “Show me failed deployments”.
• Restart services — “Restart the Traefik router” or “Redeploy the payment service”.
• Operate — deployments, compose overrides, managed networks, service backups, platform DNS, diagnostics, logs and container management (platform-owner tools).
• Read the docs — the server exposes documentation as zensailor://docs/{path} resources, so the agent answers from this manual.

Every request authenticates, binds the session to its API key, and applies the key owner's organization permissions to every tool.

Agent skill and persistent storage#

The platform ships a ready-to-paste Agent Skill — a formal SKILL.md that teaches compatible agents how to operate your PaaS: every MCP tool, the REST API syntax, permissions, and step-by-step workflows. Download it from API Documentation → Download Agent Skill inside your instance. Install it as .codex/skills/zensailor/SKILL.md, or reuse its instructions in AGENTS.md or your editor's rules file.

The skill also teaches the one rule agents must know when writing compose files: use the ${ZENSAILOR_DATA_PATH} variable for persistent volumes. ZenSailor builds in a temporary folder, so relative paths like ./data are wiped on every redeploy. The variable resolves to the service's permanent directory (/opt/zensailor/volumes/<project>/<service>) and is also available inside the container as an environment variable.

# docker-compose.yml — persistent volumes
volumes:
  - ${ZENSAILOR_DATA_PATH}/pgdata:/var/lib/postgresql/data

Configuration#

Connect your favorite AI code editor or agent with the configuration below. ZenSailor authenticates every request, binds each session to its API key, and applies the key owner's organization permissions to every tool.

[mcp_servers.zensailor]
url = "https://paas.example.com/api/mcp"
bearer_token_env_var = "ZENSAILOR_API_TOKEN"
default_tools_approval_mode = "writes"