MCP Blackboard

What it does

MCP Blackboard provides a centralized state management system for multi-agent workflows. It allows different AI agents to collaboratively store and retrieve context, plans, results, and document embeddings on a shared "blackboard," preventing the need for fragmented scratch files.

Tools

• save_plan: Stores a task execution plan in the shared state.
• mark_plan_as_completed: Updates the status of a specific plan step.
• save_result: Persists agent results for other agents to access.
• save_context_description: Maps files or URLs to descriptions in the shared state.
• get_blackboard: Retrieves the full context for a specific plan ID.
• get_plan: Fetches the current state of a plan.
• get_result: Retrieves a specific agent's output for a plan step.
• get_context: Reads and converts media content to Markdown.
• remove_stale_files: Maintenance tool to prune the local file cache.

Installation

Add the following to your claude_desktop_config.json:

{
  "mcpServers": {
    "mcp-blackboard": {
      "command": "uv",
      "args": ["--directory", "/path/to/mcp-blackboard", "run", "src/main.py"]
    }
  }
}

Supported hosts

• claude