tsdevstack
Medium

cli-mcp

@tsdevstack/cli-mcp is the Model Context Protocol server plugin for the tsdevstack CLI. It exposes 54 tools and 12 resources so AI agents like Claude Code, Cursor, and VS Code Copilot can understand and assist with your project — deployments, debugging, querying state, and following framework patterns.

Installed automatically by @tsdevstack/cli init — no separate setup needed.

How it works

The MCP server is a thin wrapper around the CLI. Each tool spawns npx tsdevstack <command> as a child process and captures the output. Resources read project files or provide hardcoded framework guides.

  • Transport: stdio (the AI client spawns the server)
  • Validation: All tool inputs validated with Zod schemas
  • Annotations: Tools include readOnlyHint, destructiveHint, and idempotentHint metadata so AI clients can make informed decisions about confirmation

Tools (54)

CategoryCountExamples
Query (read-only)13list_services, get_project_config, diff_secrets, infra_plan
Action — local14sync, generate_kong, add_service, add_bucket_storage, add_messaging_topic
Action — cloud14deploy_services, cloud_secrets_push, infra_deploy, deploy_scheduler
Action — setup & CI13cloud_init, infra_bootstrap, infra_generate, infra_generate_kong, infra_init_ci

Resources (12)

CategoryCountExamples
Project state (dynamic)8Config, infrastructure settings, secret map, Kong routes
Guides (static)4Framework guide, workflows, nest-common reference, config reference

Dynamic resources read your project files and stay current. Static guides provide framework knowledge so AI agents follow correct patterns instead of guessing.

Setup

Add to .mcp.json in your project root:

{
  "mcpServers": {
    "tsdevstack": {
      "command": "npx",
      "args": ["tsdevstack", "mcp:serve"]
    }
  }
}

For detailed tool listings, resource URIs, safety model, and client configuration, see MCP Server.