@identikey/coding-mcp

MCP server with expert AI personas for architecture reviews, code advice, and research. Runs as a tool server in Cursor, Windsurf, or any MCP-compatible client.

Quick Start

Add to your MCP config (e.g., .cursor/mcp.json):

{
  "mcpServers": {
    "coding-mcp": {
      "command": "npx",
      "args": ["-y", "@identikey/coding-mcp"],
      "env": {
        "XAI_API_KEY": "your-xai-key",
        "OPENAI_API_KEY": "your-openai-key"
      }
    }
  }
}

Or via Cursor Settings → MCP → Add Server.

Environment Variables

| Variable | Required | Description | | ---------------- | -------- | ---------------------------------------------------- | | XAI_API_KEY | Yes* | xAI/Grok API key (get one) | | OPENAI_API_KEY | Yes* | OpenAI API key (for GPT-5 reasoning) | | AI_PROVIDER | No | Default provider: xai or openai (default: xai) | | XAI_MODEL | No | xAI model override (default: grok-4.1) | | OPENAI_MODEL | No | OpenAI model override (default: gpt-5.2) |

*At least one API key required. The server auto-selects providers based on task type—xAI for quick advice, OpenAI for deep reasoning.

Using bunx

{
  "mcpServers": {
    "coding-mcp": {
      "command": "bunx",
      "args": ["@identikey/coding-mcp"],
      "env": {
        "XAI_API_KEY": "your-xai-key"
      }
    }
  }
}

Tools

ask

Natural language queries auto-routed to the best expert persona. Just ask—it figures out who should answer.

"How should I structure this React app?"  → routes to Iris (frontend)
"Is this auth implementation secure?"     → routes to Sentinel (security)
"Review this API design"                  → routes to Atlas (backend)

architect

Deep architectural review with reasoning. Pass code + task description, get structured analysis.

code-review

Git diff analysis. Point it at a repo, get review of changes vs main branch.

code-advice

Quick, focused guidance on specific problems. Lower latency than full architect review.

researcher

Multi-source research with citations. Searches across Google, arXiv, GitHub, StackExchange, etc.

persona

Direct access to a specific expert. Skip auto-routing when you know who you want.

discover

List available personas and their specialties.

screenshot

Capture screenshots from URLs for UI review.

Personas

| Persona | Focus | Style | | ------------ | ------------------------ | -------------------------------------------------- | | Charles | Pragmatic architecture | Anti-enterprise, ships fast, calls out YAGNI | | Sterling | Enterprise systems | Zero-downtime, compliance, multi-team coordination | | Ada | Algorithms & performance | Big-O, profiling, optimization | | Atlas | Backend & APIs | Contract-first, database design, caching | | Hermes | DevOps & SRE | CI/CD, observability, Kubernetes | | Sentinel | Security | Threat modeling, auth, compliance | | Iris | Frontend & UX | Accessibility, design systems, React/Vue | | Xavier | MVP development | KISS, ship it, avoid yak shaving |

Development

# Install deps
bun install

# Build
bun run build

# Test
bun test

# Run locally (for testing)
bun run start

Local MCP Config

For development, point to your local build:

{
  "mcpServers": {
    "coding-mcp-dev": {
      "command": "node",
      "args": ["/path/to/coding-mcp/build/index.js"],
      "env": {
        "XAI_API_KEY": "your-key"
      }
    }
  }
}

Project Structure

src/
├── index.ts              # MCP server entry
├── common/
│   ├── apiClient.ts      # Unified AI provider client
│   ├── providerConfig.ts # Env vars, provider selection
│   └── ...
├── personas/
│   ├── charles/          # Each persona has its own module
│   ├── sterling/
│   └── ...
└── tools/
    ├── ask.ts            # Smart routing
    ├── architect.ts      # Deep review
    └── ...

License

MIT