twining

Twining is an MCP server that provides persistent project memory for AI agents, specifically tailored to Claude Code and other MCP clients. It records decisions, findings, and knowledge as a centralized blackboard and knowledge graph, then assembles tailored context within a token budget so agents can get up-to-date information without re-explaining everything. Core commands include twining_decide for storing decisions with rationale, twining_assemble for constructing task-relevant context, and twining_why to trace the full decision chain for a given file or scope. The server ships with multiple skills and a dashboard that lets you review decisions, blackboard entries, and the knowledge graph in one place.

Prerequisites:

• Node.js installed (for npx-based MCP server)
• Access to the project repository or npm registry where twining-mcp is published

Install the MCP server using the recommended marketplace plugin flow:

# Add the marketplace (one-time)
/plugin marketplace add daveangulo/twining-mcp

# Install the plugin (includes the MCP server, 8 skills, and dashboard)
/plugin install twining@twining-marketplace

Alternatively, for MCP-only usage (non-Claude clients) you can install directly via npx:

# MCP-only install (local project setup)
claude mcp add twining -- npx -y twining-mcp --project .

Or configure via .mcp.json in your project:

{
  "mcpServers": {
    "twining": {
      "command": "npx",
      "args": ["-y", "twining-mcp", "--project", "."]
    }
  }
}

Start the MCP server (depends on your environment):

# If using the plugin setup, the server starts as part of the plugin workflow
# For MCP-only: ensure the command above is invoked by your MCP launcher

Tips and caveats:

• The dashboard is served by default at http://localhost:24282 and can be customized via TWINING_DASHBOARD_PORT.
• The Twining blackboard persists across sessions; avoid sharing sensitive project details in public repos.
• Use twining_decide to capture decisions with rationale, alternatives, confidence, and traceability to files and commits.
• The twining_assemble tool helps you build a task-specific context within a token budget, preventing information overload.
• If upgrading from manual installs, follow the migration steps in the README: remove old MCP config, install the plugin, and keep the .twining/ directory to preserve state.
• The knowledge graph visualizes files, decisions, and their relationships; use twining_what_changed to see changes since a given point in time.