🚀 Gemini CLI — Rocket.Chat MCP Server Generator

A production-grade CLI that dynamically generates scoped, AI-ready Model Context Protocol (MCP) servers from the Rocket.Chat OpenAPI specification — powered by a Dynamic AST Scaffolding Engine.

🎬 Demo

https://github.com/user-attachments/assets/07d8e46c-3b32-41d5-901c-8c6ef9cc5483

✨ What Is This?

This tool lets you describe a Rocket.Chat use case in plain English and instantly receive a fully compiled, production-ready MCP server that exposes only the API capabilities relevant to your workflow. The generated server can be plugged directly into any MCP host — Claude Desktop, Cursor, or any custom agent.

"I want to kick a user from a channel"
         ↓  Gemini semantic analysis
    Capabilities: channels_kick, users_info
         ↓  Dynamic AST Scaffolding Engine
    Generated: out-mcp/build/server.js
         ↓  One command
    Running in Claude Desktop

🏗️ Architecture

gemini-cli-rocketchat-mcp/
├── src/
│   ├── index.ts              # CLI entry point (generate, chat, install commands)
│   ├── generator/
│   │   └── mcp.ts            # Dynamic AST Scaffolding Engine
│   ├── parser/
│   │   └── openapi.ts        # OpenAPI $ref resolver (swagger-parser)
│   └── semantic/
│       └── search.ts         # Intent → workflow mapping
├── scripts/
│   └── merge-openapi.ts      # Merges Rocket.Chat OpenAPI spec files
├── merged-openapi.json       # Pre-merged Rocket.Chat OpenAPI spec
└── out-mcp/                  # Generated MCP server (git-ignored)
    └── build/server.js

Dynamic AST Engine

Instead of hardcoded tool definitions, the generator:

Uses @apidevtools/swagger-parser to fully dereference all $ref pointers in the OpenAPI spec
Walks the AST to extract endpoint schemas for only the requested workflows
Scaffolds a complete, type-safe MCP server with accurate JSON Schema tool definitions
Auto-builds the output with npm install && tsc

📋 Prerequisites

Node.js 18+
A running Rocket.Chat instance (local Docker or remote)
Rocket.Chat API credentials (User ID + Auth Token)

Get Your Credentials

Admin → My Account → Personal Access Tokens → Create Token

🚀 Quick Start

1. Install

git clone https://github.com/<your-username>/gemini-cli-rocketchat-mcp
cd gemini-cli-rocketchat-mcp
npm install
npm run build
npm link   # Makes 'gemini' available globally

2. Generate a Scoped MCP Server

gemini generate -i

You'll be prompted:

✨ Welcome to the Gemini AI Rocket.Chat Server Generator ✨

Describe your use case: I want to kick a user from a channel

🤖 Gemini recommends: kickUser, onboardUser

Proceed? (Y/n): y

✅ Build complete! Server is fully compiled and ready to test.

3. Test Your Server

export RC_USER_ID="your_user_id"
export RC_AUTH_TOKEN="your_auth_token"

gemini chat \
  --mcp "node out-mcp/build/server.js" \
  --prompt "kick rocket.cat from general"

4. Install into Claude Desktop

gemini install
# → Writes to ~/AppData/Roaming/Claude/claude_desktop_config.json
# → Restart Claude Desktop → 🔨 hammer icon confirms tools are loaded

📖 CLI Commands

`gemini generate`

Generates a scoped MCP server from the Rocket.Chat OpenAPI spec.

gemini generate [options]

Options:
  -s, --spec <file>       Path to OpenAPI spec JSON/YAML  [default: merged-openapi.json]
  -w, --workflows <list>  Comma-separated workflow names  [default: provisionChannel,broadcastMessage]
  -o, --outDir <dir>      Output directory                [default: ./out-mcp]
  -i, --interactive       AI-guided interactive mode

Interactive mode (-i) lets Gemini analyze your plain-English description and recommend the best matching Rocket.Chat workflows automatically.

`gemini chat`

Tests the generated MCP server using a real MCP client with natural language prompt routing.

gemini chat --mcp "<command>" --prompt "<text>"

# Example:
gemini chat \
  --mcp "node out-mcp/build/server.js" \
  --prompt "kick rocket.cat from the general channel"

The client:

Spawns the MCP server process (with full env var inheritance)
Performs the complete MCP handshake (initialize → capabilities)
Routes the prompt to the matched tool via keyword analysis
Calls the tool and displays the live Rocket.Chat API response

`gemini install`

Installs the generated server into Claude Desktop's config file.

gemini install [options]

Options:
  --mcp-dir <dir>         Path to out-mcp directory  [default: ./out-mcp]
  --rc-url <url>          Rocket.Chat URL            [default: http://localhost:3000]
  --rc-user-id <id>       RC User ID                 [default: $RC_USER_ID env var]
  --rc-auth-token <token> RC Auth Token              [default: $RC_AUTH_TOKEN env var]

🔐 Environment Variables

| Variable | Description | Required | |---|---|---| | RC_URL | Rocket.Chat server URL | No (default: http://localhost:3000) | | RC_USER_ID | Your Rocket.Chat User ID | Yes | | RC_AUTH_TOKEN | Your Rocket.Chat Auth Token | Yes |

Set them before any command:

export RC_USER_ID="bebcLTPed56WAFbZW"
export RC_AUTH_TOKEN="your_token_here"

🧪 Supported Workflows

| Workflow ID | Description | Tools Generated | |---|---|---| | kickUser | Remove a user from a channel | channels_kick | | onboardUser | Create and configure a new user | users_create | | inviteUser | Invite a user to a channel | channels_invite | | provisionChannel | Create and set up a new channel | channels_create | | broadcastMessage | Post a message to a channel | chat_postMessage |

Note: The generator dynamically reads these from the OpenAPI AST so new endpoints added to the Rocket.Chat API are automatically available.

🖥️ Using with Claude Desktop

After running gemini install, restart Claude Desktop. The 🔨 hammer icon in the chat input confirms your Rocket.Chat tools are loaded.

Example prompts you can use:

"Kick rocket.cat from the general channel" → channels_kick
"Create a new user named John with email john@example.com" → users_create
"Invite alice to the #announcements channel" → channels_invite

🧩 How the Generated Server Works

The generated out-mcp/build/server.js:

Runs as a stdio-based MCP server (compatible with all MCP hosts)
Reads credentials from environment variables at startup
Exposes tools derived from your selected workflows with accurate JSON Schema
Routes tool calls through a universal fetch router to the Rocket.Chat REST API
Returns structured responses for both success and error cases

🔮 Roadmap (GSoC 2026)

[ ] Phase 1: Full OpenAPI AST coverage (all 300+ Rocket.Chat endpoints)
[ ] Phase 2: Name-to-ID resolver tools (channels_list, users_info) for seamless natural language operation
[ ] Phase 3: LLM schema bridge — native Gemini FunctionDeclaration generation from MCP JSON Schemas
[ ] Phase 4: Multi-step agentic workflows (e.g. "onboard a new team member" → create user → add to channels → send welcome message)

📄 License

MIT

MCP Servers