AgentPM

What is AgentPM?

AgentPM is a planning and orchestration system for AI-driven software development. It integrates as a local MCP server with any IDE that supports Anthropic’s Model Context Protocol (Cursor, Augment, VS Code Copilot, Cline, Roo), serving the role of a product manager to help developers plan, prioritize, and execute complex projects.

How to use AgentPM?

Install by running npx -y @gannonh/agent-pm@latest and configure the MCP server in your IDE’s settings (e.g., .cursor/mcp.json). Required environment variables: PROJECT_ROOT and ANTHROPIC_API_KEY. Optionally set PERPLEXITY_API_KEY for research-backed generation. The server exposes tools such as apm_task for querying and managing tasks.

Key features of AgentPM

• Frictionless setup – start by chatting with your coding agent.
• Token/context optimization with dynamic, economical tools.
• Intelligent context management – delivers right information at the right time.
• Structured markdown output – no JSON or plain text to decode.
• Integrated documentation retrieval via Context7.
• AI-powered generation using Claude Sonnet 3.7 and optional Perplexity API.

Use cases of AgentPM

• Planning and breaking down complex software projects into actionable tasks with dependencies.
• Guiding technical decisions and system design during implementation.
• Automatically maintaining living documentation that evolves with the project.
• Integrating best practices like TDD and vertical slicing into the development workflow.

FAQ from AgentPM

How does AgentPM differ from plain MCP servers or memory banks?

AgentPM consolidates functionality into a core set of dynamically chosen tools that are contextually economical and easy for coding agents to use. It eliminates the need for manual context handling or separate “memory banks,” delivering relevant documentation automatically.

What are the runtime dependencies?

Node.js version 20.0.0 or higher, an Anthropic API key (required), and optionally a Perplexity API key for research-backed task generation. The server runs locally and uses Claude Sonnet 3.7 for task generation.

Where does AgentPM store its data?

All artifacts are stored locally in the project directory. The default artifacts directory is apm-artifacts/ and the project brief file default is project-brief.md. File paths and names can be configured via environment variables.

What transport does AgentPM use?

AgentPM is an MCP server; it communicates with IDEs that support the Model Context Protocol. Configuration examples use stdio (via npx). No authentication beyond the API keys is required for the MCP transport itself.

Are there any known limits or debug options?

Default maximum tokens for Claude API calls is 64,000; for Perplexity it is 1,024. Debug logging can be enabled with DEBUG_LOGS=true to write detailed logs to a logs directory. Most configuration options are tuned for optimal results and should be left at defaults unless specific needs arise.