@dici1435/spec-review-mcp

LLM-powered spec compliance review MCP server for dici-spec projects. Reviews code against spec documents and constraints, returning structured findings with category, severity, file location, and description.

Installation

npm install -g @dici1435/spec-review-mcp
# or run directly
npx @dici1435/spec-review-mcp

Added to a project via dici-spec add spec-review, which prompts for your LLM API key.

Tools

| Tool | Description | |---|---| | review_code | Review code files against spec documents for compliance. Accepts spec paths, code paths, and optional constraints. Returns structured findings. |

Finding Structure

Each finding includes:

| Field | Values | |---|---| | category | spec-compliance, missing-implementation, deviation, architecture, correctness | | severity | blocker, major, minor, nit | | file | Path to the file with the finding | | line | Line number (when applicable) | | description | Detailed explanation of the finding |

Environment Variables

| Variable | Required | Default | Description | |---|---|---|---| | LLM_API_KEY | Yes | — | API key for the LLM provider (OpenAI or Anthropic) | | LLM_BASE_URL | No | https://api.openai.com/v1 | Base URL for the LLM API. Set to https://api.anthropic.com/v1 to use Anthropic. | | LLM_MODEL | No | gpt-4o (OpenAI) / claude-3-5-sonnet-20241022 (Anthropic) | Model to use for review | | PROJECT_ROOT | No | process.cwd() | Root directory of the project | | SPECS_DIR | No | agents/specs | Directory containing spec files (relative to PROJECT_ROOT) | | CONSTRAINTS_DIR | No | agents/constraints | Directory containing constraint files (relative to PROJECT_ROOT) |

The provider is auto-detected from LLM_BASE_URL — no separate provider flag needed. If the URL contains anthropic.com, the Anthropic Messages API is used. Everything else uses the OpenAI Chat Completions API.

Provider Configuration

OpenAI (default)

"dici-spec-review": {
  "command": "npx",
  "args": ["@dici1435/spec-review-mcp"],
  "env": {
    "LLM_API_KEY": "sk-...",
    "LLM_MODEL": "gpt-4o"
  }
}

Anthropic

"dici-spec-review": {
  "command": "npx",
  "args": ["@dici1435/spec-review-mcp"],
  "env": {
    "LLM_API_KEY": "sk-ant-...",
    "LLM_BASE_URL": "https://api.anthropic.com/v1",
    "LLM_MODEL": "claude-sonnet-4-5"
  }
}

The provider is detected automatically from LLM_BASE_URL. When using Anthropic, the server uses the /v1/messages endpoint with x-api-key authentication and passes the system prompt as Anthropic's top-level system parameter.

OpenAI-compatible providers (OpenRouter, Azure, etc.)

Any provider with an OpenAI-compatible /chat/completions endpoint works by setting LLM_BASE_URL:

"LLM_BASE_URL": "https://openrouter.ai/api/v1",
"LLM_MODEL": "anthropic/claude-sonnet-4-5"

How It Works

1. Loads spec documents from SPECS_DIR and constraints from CONSTRAINTS_DIR
2. Reads the specified code files
3. Assembles a review prompt with system instructions, spec content, code content, and constraint context
4. Sends the prompt to the configured LLM (auto-detecting OpenAI vs. Anthropic from LLM_BASE_URL)
5. Parses the response into structured findings

Transport

Stdio (stdin/stdout). Designed to be launched by AI agent IDEs as a child process.

License

MIT