MCP Setup

Configure required MCP servers for agent-harness token efficiency and learning loops.

Required MCP Servers

Setup Instructions

Quick setup - Run scripts/setup-all.sh in your project directory:

cd /path/to/your/project
bash ~/.claude/skills/mcp-setup/scripts/setup-all.sh

This will:

1. Create mcp/ folder in your project
2. Clone and build token-efficient MCP
3. Clone and install context-graph MCP
4. Prompt for Voyage AI API key
5. Generate .mcp.json in project root

Manual setup (if needed):

1. Install token-efficient MCP in mcp/

mkdir -p mcp
git clone https://github.com/gurusharan/token-efficient-mcp.git mcp/token-efficient-mcp
cd mcp/token-efficient-mcp
npm install
npm run build

2. Install context-graph MCP in mcp/

git clone https://github.com/gurusharan/agent-harness.git mcp/context-graph-mcp
cd mcp/context-graph-mcp/context-graph-mcp  # or just mcp/context-graph-mcp
pip install -r requirements.txt

3. Create .mcp.json in project root

{
  "mcpServers": {
    "token-efficient": {
      "command": "node",
      "args": ["mcp/token-efficient-mcp/dist/index.js"]
    },
    "context-graph": {
      "command": "uv",
      "args": ["--directory", "mcp/context-graph-mcp", "run", "python", "server.py"],
      "env": {
        "VOYAGE_API_KEY": "your_key_here"
      }
    }
  }
}

4. Restart Claude Code

After setup, restart Claude Code to load MCP servers.

Verification

After setup, test:

# Via context-graph MCP
context_store_trace(decision="Test setup", category="general")
context_list_categories()

Troubleshooting

Scripts