AI-Prompt-Guide

AI Prompt Guide MCP is a server that orchestrates multiple AI agents through structured, linked Markdown documents. It enables you to define specifications, guides, and architecture decisions, then assign tasks to a coordinator or subagents. The server handles automatic context injection from linked content using @references and manages workflow metadata to drive multi-agent collaboration. You can access a library of predefined workflows via the get_workflow tool or use the Claude Code plugin to trigger workflows directly from your editor or IDE. This makes it possible to drive complex multi-agent pipelines with a single command, while keeping the coordinator separated from implementation details for unbiased reviews.

Typical usage involves writing or linking documents that describe a feature or project, then choosing an appropriate workflow (planning, development, auditing, etc.). The MCP server loads relevant content automatically when tasks start, injecting context so agents can reason about dependencies, requirements, and constraints without manual copy-paste. As tasks are completed, the coordinator can review changes and archive completed work, maintaining a clear record of decisions and outcomes across sessions.

Prerequisites:

• Node.js 18+ installed on your system
• pnpm 10.x installed

Zero-config (recommended):

1. Clone and run immediately without a separate install:

git clone https://github.com/your-org/AI-Prompt-Guide-MCP.git
cd AI-Prompt-Guide-MCP
pnpm start

If you plan to develop or customize tooling: 2) Install development tools and build steps:

./scripts/dev-mode-on.sh
pnpm install
pnpm build

3. Run in development or production mode as needed:

pnpm dev      # Development server
pnpm start    # Production server
pnpm inspector # Run with inspector for debugging

Configuring MCP client integration: 4) If you are integrating with an MCP client directly (non-Claude Code), add or update your MCP config:

{
  "mcpServers": {
    "ai-prompt-guide-mcp": {
      "command": "node",
      "args": ["/path/to/AI-Prompt-Guide-MCP/dist/index.js"],
      "env": {
        "MCP_WORKSPACE_PATH": "/custom/workspace/path"
      }
    }
  }
}

Note: No build step is required if you use the provided production bundle; replace /path/to/ with your actual clone location.

Tips and common considerations:

• MCP_WORKSPACE_PATH controls where the server stores documents, tasks, and archives; adjust as needed for your project structure.
• DOCS_BASE_PATH, ARCHIVED_BASE_PATH, and COORDINATOR_BASE_PATH follow the default zero-config layout unless overridden by per-project .mcp-config.json.
• REFERENCE_EXTRACTION_DEPTH controls how deep @references are resolved (range 1-5; default is 3).
• LOG_LEVEL can be adjusted to debug, info, warn, or error to tune verbosity.
• When using Claude Code plugin, you can invoke workflows via both plugin commands and get_workflow calls; this ensures parity between plugin-based and MCP-server-based usage.
• For per-project configuration, you can supply a .mcp-config.json in the project root to override default paths and environment variables.
• The server supports automatic archiving of coordinator tasks upon completion to keep project history clean.