MCP Server

MCP Server

skene-growth provides an MCP (Model Context Protocol) server that exposes codebase analysis capabilities to AI assistants like Claude Desktop and Claude Code.

The server communicates via stdio and exposes 12 tools organized into tiers by speed and complexity. Tier 1 tools run in under a second with no LLM calls. Tier 2 tools perform LLM-powered analysis with automatic caching. Tier 3 tools combine cached results into final outputs.

Installation

Install with the mcp optional dependency:

# With pip
pip install skene-growth[mcp]

# With uv
uv pip install skene-growth[mcp]

Or run directly with uvx (no installation required):

uvx --from "skene-growth[mcp]" skene-growth-mcp

Configuration

Claude Desktop

Add to your Claude Desktop config file:

• macOS: ~/Library/Application Support/Claude/claude_desktop_config.json
• Windows: %APPDATA%\Claude\claude_desktop_config.json

Option 1: Installed locally

{
  "mcpServers": {
    "skene-growth": {
      "command": "skene-growth-mcp",
      "env": {
        "SKENE_API_KEY": "your-api-key"
      }
    }
  }
}

Option 2: Via uvx (no install needed)

{
  "mcpServers": {
    "skene-growth": {
      "command": "uvx",
      "args": ["--from", "skene-growth[mcp]", "skene-growth-mcp"],
      "env": {
        "SKENE_API_KEY": "your-api-key"
      }
    }
  }
}

Claude Code

Add to .mcp.json in your project root or ~/.claude/settings.json for global configuration:

{
  "mcpServers": {
    "skene-growth": {
      "command": "skene-growth-mcp",
      "env": {
        "SKENE_API_KEY": "your-api-key"
      }
    }
  }
}

The uvx variant works identically -- use "command": "uvx" with "args": ["--from", "skene-growth[mcp]", "skene-growth-mcp"].

Available Tools

Tier 1: Quick Tools (< 1s, no LLM)

These tools run instantly and require no LLM provider configuration.

Tier 2: Analysis Tools (5-15s, uses LLM, cached)

These tools call the configured LLM provider and cache results per phase. Each tool accepts path (required) and force_refresh (optional boolean, default false).

Tier 3: Generation Tools (5-15s)

These tools combine cached analysis results into final outputs.

Utility Tools

Typical Workflow

The tools are designed to be called in a specific sequence. Tier 2 analysis tools populate a cache, and Tier 3 generation tools read from that cache.

1. get_codebase_overview        -- Understand project structure
2. analyze_tech_stack           -- Detect technologies (cached)
3. analyze_growth_hubs          -- Find growth features (cached)
4. generate_manifest            -- Combine into GrowthManifest (reads cache)
5. generate_growth_template     -- Create PLG template (optional)
6. write_analysis_outputs       -- Save files to ./skene-context/

For a full analysis with product documentation, add analyze_product_overview and analyze_features before generate_manifest, and pass product_docs=true to both generate_manifest and write_analysis_outputs.

Cache Configuration

The MCP server caches Tier 2 analysis results to avoid redundant LLM calls. The cache uses a two-layer strategy:

• Memory cache: In-process dictionary, fastest lookup
• Disk cache: JSON files in the cache directory, persists across server restarts

Invalidation

Cache entries are automatically invalidated when:

• The TTL expires (default: 1 hour)
• Marker file hashes change -- any modification to dependency/config files triggers invalidation. Tracked files include: package.json, requirements.txt, pyproject.toml, Cargo.toml, go.mod, Gemfile, composer.json, and their lockfiles.
• Source directory mtimes change -- modifications to common source directories (src, lib, app, pages, components, api, server, client) trigger invalidation.

You can also manually invalidate with the clear_cache tool or by passing force_refresh=true to any analysis tool.

Phase-Specific Keys

Each analysis phase has its own independent cache entry. Running analyze_tech_stack does not invalidate the cache for analyze_growth_hubs. The phases are: tech_stack, product_overview, current_growth_features, features, industry, manifest, growth_template.

Environment Variables

Using Local LLMs

For LM Studio or Ollama, no API key is needed. Set SKENE_PROVIDER in the env block:

{
  "mcpServers": {
    "skene-growth": {
      "command": "skene-growth-mcp",
      "env": {
        "SKENE_PROVIDER": "lmstudio",
        "SKENE_MODEL": "your-loaded-model"
      }
    }
  }
}

For Ollama, use "SKENE_PROVIDER": "ollama". The server connects to the default local endpoint for each provider.

See LLM Providers for details on configuring each provider.

Running Manually

You can start the MCP server directly for testing or debugging:

# Via the entry point
skene-growth-mcp

# Via Python module
python -m skene_growth.mcp

The server communicates via stdio (standard input/output) as required by the MCP protocol. It is not meant to be run interactively -- it expects JSON-RPC messages on stdin and writes responses to stdout.

Next Steps

• Configuration -- config file options and environment variable priority
• LLM Providers -- setup for OpenAI, Gemini, Claude, LM Studio, Ollama
• CLI Reference -- the full CLI if you prefer command-line usage over MCP