11 CLI Tools That Are Also MCP Servers

Every tool in the DevForge suite doubles as an MCP server. AI coding agents call them directly � zero glue code.

May 19, 2026 8 min read

The Model Context Protocol is changing how AI coding agents interact with developer tools. Instead of copy-pasting terminal output or writing custom API integrations, agents like Claude Code and Cursor can now call CLI tools directly through MCP.

But here's the problem: most CLI tools don't speak MCP. You need wrapper scripts, adapter layers, or API gateways. That's extra code to maintain, extra points of failure, and extra time between "I want my agent to use this tool" and "it actually works."

We solved this differently. Every tool in the DevForge suite has a built-in mcp subcommand. One command turns any of our 11 CLI tools into an MCP server over stdio. No wrappers. No adapters. No glue code.

How It Works

Each DevForge tool is built with click-to-mcp, our open-source library that auto-wraps Click and Typer CLIs as MCP servers. Every subcommand becomes an MCP tool. Every flag becomes a parameter. The AI agent discovers available tools, calls them, and reads structured output.

# Start any tool as an MCP server
deploydiff mcp
envault mcp
configdrift mcp
apighost mcp
# ... and 7 more

In your AI agent's MCP config, add entries like:

{
  "mcpServers": {
    "deploydiff": {
      "command": "deploydiff",
      "args": ["mcp"]
    },
    "envault": {
      "command": "envault",
      "args": ["mcp"]
    }
  }
}

That's it. The agent now has access to deploydiff preview, deploydiff cost, envault sync, and every other subcommand � as native MCP tools.

All 11 Tools at a Glance

Why This Matters

Zero Integration Tax

Most MCP servers are built from scratch for a specific tool. That means each one needs its own transport layer, error handling, schema definition, and test suite. With click-to-mcp, the CLI is the MCP server. The same codebase serves humans in the terminal and AI agents over stdio.

Consistent DX Across 11 Tools

If you know how to use one DevForge tool as an MCP server, you know how to use all of them. Same mcp subcommand. Same config pattern. Same output format. No per-tool documentation hunting.

Real Workflows, Not Demos

Here's what an AI agent workflow looks like with these tools connected:

1. Pre-deploy safety net: Agent calls deploydiff preview on a Terraform plan, sees destructive changes, calls deploydiff cost to check the monthly impact, then calls deploydiff rollback to prepare recovery commands.
2. API evolution: Agent calls api-contract-guardian check on an OpenAPI spec diff, finds a breaking change, calls api-contract-guardian migrate to generate a migration guide, then opens a PR with both.
3. Environment sync: Agent calls configdrift detect to compare staging and production configs, finds drift in 3 services, calls envault sync to align them, then verifies with another drift check.
4. Dead code cleanup: Agent calls deadcode scan on a codebase, gets a report of unused exports, removes them, runs the test suite to verify nothing breaks.
5. Test data seeding: Agent calls json2sql convert on fixture files to seed a test database, then calls apighost serve to spin up a mock API for integration tests.

Getting Started

# Install any tool
pip install deploydiff envault configdrift

# Verify MCP mode works
deploydiff mcp  # starts stdio MCP server

# Add to your AI agent's MCP config
# See above for config format

The Bigger Picture

MCP is still early. Most developer tools aren't MCP-aware. By making every tool in our suite an MCP server out of the box, we're betting that AI agents will become the primary interface for developer infrastructure � and that tools which speak MCP natively will have a structural advantage.

If you're building CLI tools, check out click-to-mcp on GitHub. It's open source, works with any Click or Typer CLI, and takes about 2 lines of code to integrate.