AI Agent Team

What ai-agent-team MCP server does

ai-agent-team combines Claude Code agent roles with a Thread Manager MCP server and several separately installable skills. The agent roles cover product planning, frontend implementation, backend services, full-stack delivery, testing, DevOps, and technical leadership. Their slash commands provide focused entry points such as /pm, /fe, /be, /fs, /qa, /ops, and /tl.

Thread Manager is the central persistence layer. It keeps task conversations available across sessions, supports multiple concurrent threads, restores context when a thread is selected, and provides semantic retrieval over previous messages. Its Git integration can create task branches, track file changes, and provide code statistics.

Additional skills cover changelog generation, software copyright application materials, desktop file organization, and visual note creation. The changelog tool can analyze Git history, produce Markdown or HTML output, update an existing changelog, and create a GitHub Release. Other skills generate HTML and PNG visual notes, organize files with dry-run support, or prepare source and explanatory documents for copyright applications.

How it works

The MCP component runs as a Node.js process from the installed package. Claude Code sends Thread Manager requests to it, and the manager reads or writes conversation data in a SQLite database. An embedded Xenova/all-MiniLM-L6-v2 model generates local embeddings for message storage and semantic search, so the memory workflow can operate without an external embedding service.

Threads can be listed with /threads, inspected with /thread info, or switched with /thread switch <id>. Role-specific startup commands such as /pm-start, /fe-start, /be-start, /qa-start, and /fs-start create focused work threads. The project also includes shell and PowerShell CLI wrappers for invoking agent roles directly.

Setup and configuration

Install the package globally, initialize its Claude configuration, then register the Thread Manager entry point with Claude Code:

npm install -g ai-agent-team
ai-agent-team init
claude mcp add thread-manager node "/path/to/.claude/skills/thread-manager/dist/index.js"

Initialization can be global, placing configuration under ~/.claude/, or project-local, placing it under ./.claude/. Claude Code and Git are listed as system requirements. After registering the MCP process, restart Claude Code and verify it with /threads, /thread info, or /pm-start "test".

The independent agent skills can be installed into projects with the skills install command. Examples include npx skills install peterfei/ai-agent-tech-leader and the corresponding repositories for the product manager, frontend, backend, QA, DevOps, and full-stack roles.

Tools and capabilities

• Persistent task threads with context restoration and parallel task handling.
• Semantic search across saved messages using a locally bundled embedding model.
• Git task branches, file-change tracking, and code statistics.
• Specialist commands for product, frontend, backend, full-stack, QA, DevOps, and architecture work.
• Changelog generation with Markdown or HTML output and GitHub Release support.
• Desktop organization with categorization, version-aware deduplication, and dry-run planning.
• Visual note generation with five built-in presentation styles and HTML/PNG output.
• Software copyright document generation with source-code formatting and PDF-oriented output.

Limitations and notes

Thread Manager must be explicitly registered with Claude Code; initialization alone does not enable its thread commands. The embedded model increases the installed package size and supports offline embedding generation. Global and project-local configurations can coexist, with the project-local configuration taking precedence when present. The role skills and utility skills are separate components, so installing the package does not replace the documented skills install steps for those independent repositories.