tech-lead-stack

The MCP server for the Tech-Lead Stack: agent skills and workflows for tech leads, product managers and HR, served to any MCP-capable editor or agent. The skills are bundled; nothing else needs to be cloned.

Requires Node.js 22.5 or later.

Set it up

npx -y tech-lead-stack@latest init

It finds the editors on this computer (Claude Code, Claude Desktop, Cursor, Continue, Cline, Gemini), shows what it will change, and asks once. It adds the toolbox to each editor, or puts it behind a gateway such as slm-gate if you already use one, so no tool is listed twice. It never replaces a connection you already have. It also installs:

• Claude Code's /tls:* slash commands;
• Cursor skills and Continue prompts;
• the workflow files, into the project you run it from (Antigravity and Gemini read them there);
• RTK, which cuts the tokens your assistant spends reading command output. It is a checked download of a tested version, turned on for Claude Code.

Then it creates your settings file and, if you like, asks for a database address and API keys, which are not shown as you type.

--dry-run only shows the plan. --yes accepts the recommended choices without questions, for AI assistants and scripts. --no-rtk and --no-project skip those parts. Restart your editors afterwards. The copied files then stay current by themselves: each time the server starts on a new release, it updates them (new commands appear from the next session). Any you edited are left alone; TLS_AUTO_REFRESH=0 turns this off.

To remove it all: npx -y tech-lead-stack@latest uninstall shows what it would remove, and --apply removes it. Your settings file and RTK are kept.

Connect it by hand

Claude Code

claude mcp add tech-lead-stack -- npx -y tech-lead-stack

Cursor, Continue, Cline, Antigravity and other MCP clients

Add this to the client's MCP server configuration:

{
  "mcpServers": {
    "tech-lead-stack": {
      "command": "npx",
      "args": ["-y", "tech-lead-stack"]
    }
  }
}

Then ask your agent to list the available skills.

Behind a gateway such as slm-gate

If your editor already talks to a gateway, let the gateway start this server instead of registering it a second time. For slm-gate, add these two values to the env block of your existing slm-gate entry, then restart your editor:

"TLS_ADAPTER": "on",
"DOWNSTREAM_MCP": "{\"command\":\"npx\",\"args\":[\"-y\",\"tech-lead-stack@2\"]}"

@2 pins the current major version; init writes the right one for you.

The tools then appear under the gateway's name, such as mcp__slm-gate__list_skills. See Running the stack behind an upstream MCP proxy.

Check your setup

npx -y tech-lead-stack@latest doctor

It checks Node.js, your settings file, which features your keys turn on, the usage-metrics database, and how each editor on this computer reaches the toolbox, then says what to do about anything missing. It only reads; nothing is changed. Add --json to get the same report for an AI assistant.

Settings and API keys

Put settings in one file, ~/.tech-lead-stack/.env, and every editor uses them. A value in an editor's MCP env block wins over the file, for that editor only. The server never reads the .env of the project you have open; that file belongs to your app.

Tools that call models themselves, such as the reflexion loops, need the matching API keys (for example ANTHROPIC_API_KEY, GEMINI_API_KEY). See Configuration.

Usage metrics

Add DATABASE_URL to ~/.tech-lead-stack/.env and every skill and tool call is recorded. To see the results, run the web app against the same database: Run it on your own machine. Without DATABASE_URL, nothing is recorded and everything else works.

Verify the package

Every release is built and published by GitHub Actions with npm provenance:

npm audit signatures

Help, bugs and security

• Questions and bugs: SUPPORT.md
• Vulnerabilities (report privately): SECURITY.md

Licence

MIT