MCP.so
Sign In
A

Agentteam

@RichardLemmon

About Agentteam

A reusable AI software development team built on MCP. 13 specialized agents (Project Manager, Backend, Frontend, QA, Security, DevOps, UX, and more) collaborate via shared SQLite state. Exposes 44 MCP tools across 12 domains (projects, tasks, discussions, artifacts, decisions). O

Config

Add this server to your MCP-compatible client using the configuration below.

{
  "mcpServers": {
    "agent-team": {
      "command": "npx",
      "args": [
        "agent-team-mcp"
      ]
    }
  }
}

Tools

46

Create a new project with a name and description

Get a project by ID

Update the status of a project (active → paused → completed or archived)

List all projects, optionally filtered by status

Permanently delete a project and ALL associated data (tasks, work entries, discussions, artifacts, etc.). This is irreversible.

Get the latest summary for a project

Create a new versioned summary for a project

Get a specific summary version by its ID

List all summary versions for a project

Add a team member to a project with a role

Remove a team member from a project (soft delete)

List team members for a project, optionally including removed members

Create a new task in a project

Update a task's status, description, or assignee

Get a task by ID

List tasks for a project, optionally filtered by assignee or status

Log a work entry for a task by a team member

Get work entries for a team member, optionally filtered by task

Get all work entries for a member within a project

Add a comment to a task

List all comments on a task

List all comments made by a member within a project

Create a new discussion thread in a project

Add a participant to an existing discussion

Post a message to a discussion (participant must already be in the discussion)

Update the summary of a discussion

Get a discussion including its participants and messages

List discussions in a project, optionally filtered by participant

Log a decision made within a project

List all decisions for a project

Get a decision by ID

Returns the shared team protocol, constraints, and efficiency rules that all specialist agents must follow. Call this once on startup.

Returns step-by-step instructions for how to orchestrate an AgentTeam. Call this FIRST when a user asks to spin up a team, build something with the team, or use AgentTeam. The instructions explain the full pipeline: spawning the PM, parsing the dispatch manifest, launching specialists, handling user questions and expansion requests.

Returns the prompt file for a specific agent role. Use this to load agent identity prompts before spawning specialists. Valid roles: project-manager, product-manager, ux-ui-designer, ux-researcher, frontend-developer, backend-developer, full-stack-developer, mobile-developer, devops-engineer, qa-engineer, security-engineer, data-engineer, data-scientist.

Share an artifact (document, code, etc.) within a project. Research artifacts must use structured JSON: { "summary": "one sentence", "findings": [{ "claim": "...", "evidence": "url or source", "confidence": "high|medium|low" }], "recommendations": ["..."], "blockers": ["..."], "open_questions": ["..."] }. Code artifacts are exempt — use the appropriate file format.

Update the content or title of a shared artifact

List shared artifacts in a project, optionally filtered by type

Get a shared artifact by ID

Log a user-facing journal entry — captures decisions, preferences, and reasoning from conversations that would otherwise be lost. project_id is optional; omit it for general cross-project conversations.

List journal entries in chronological order. Optionally filter by project_id; omit to list all entries across all projects.

Log a question for the user. The orchestrating skill will surface it after dispatch. Include context about why this question matters or what is blocked.

List questions logged by specialists for the user. Filter by status (pending, answered) to find unanswered questions.

Write the user's answer to a previously asked question

Request additional team members when your assigned work grows beyond expected scope. The PM will evaluate and approve or deny.

List team expansion requests for a project, optionally filtered by status (pending, approved, denied)

Approve or deny a team expansion request (PM only)

Overview

What is AgentTeam?

AgentTeam is a reusable AI software development team built on the Model Context Protocol (MCP). Thirteen specialized agents — Product Manager, Project Manager, UX Researcher, UX/UI Designer, Frontend, Backend, Full-Stack, Mobile, DevOps, QA, Security, Data Engineer, and Data Scientist — collaborate on software projects through a shared SQLite database, each strictly constrained to their role. It is designed to be called from any Claude Code project via MCP.

How to use AgentTeam?

Install the MCP server with one command: claude mcp add agent-team -- npx agent-team-mcp. Alternatively, edit your MCP config manually. The /team skill is installed automatically on first startup. Invoke with /team build me a REST API for task management or /team --projects to manage existing projects.

Key features of AgentTeam

  • 44 MCP tools across 12 domains (projects, tasks, discussions, etc.)
  • Thirteen role-constrained AI agents collaborating in parallel
  • Shared SQLite database persists all project state across sessions
  • Project lifecycle: active → paused → archived → closed
  • User journal captures decisions, preferences, and reasoning
  • Token-efficient architecture with lazy-loaded team protocol

Use cases of AgentTeam

  • Spin up a full product team to design and build a new feature
  • Investigate a technical subject with specialists producing structured research artifacts
  • Resume paused software projects without losing context or prior decisions
  • Automate multi-agent code generation, review, and deployment planning

FAQ from AgentTeam

What runtime environment does AgentTeam require?

AgentTeam runs as a Node.js MCP server and is installed via npx, so Node.js is required. The server uses SQLite for persistence and communicates over stdio.

How do I install AgentTeam?

Install it with claude mcp add agent-team -- npx agent-team-mcp. You can also manually add it to claude_desktop_config.json or run from a local clone after npm install and npm run build.

What is the user journal?

The user journal is a persistent log of your decisions, preferences, and reasoning captured during team conversations. It is scoped to a project (or global) and can be queried by future agents via list_journal_entries.

Can I reuse the same team across sessions?

Yes. All project state is persisted in a SQLite database. Any agent can rejoin a project by reading the current summary and picking up where the team left off. Projects support lifecycle transitions (active, paused, archived, closed).

How does AgentTeam avoid redundant work?

Specialists share structured research artifacts via the share_artifact tool, and the Project Manager orchestrates tasks so no agent re-researches what another has already found. The team protocol is lazy-loaded to save ~6,000 words of context.

Frequently asked questions

What runtime environment does AgentTeam require?

AgentTeam runs as a Node.js MCP server and is installed via `npx`, so Node.js is required. The server uses SQLite for persistence and communicates over stdio.

How do I install AgentTeam?

Install it with `claude mcp add agent-team -- npx agent-team-mcp`. You can also manually add it to `claude_desktop_config.json` or run from a local clone after `npm install` and `npm run build`.

What is the user journal?

The user journal is a persistent log of your decisions, preferences, and reasoning captured during team conversations. It is scoped to a project (or global) and can be queried by future agents via `list_journal_entries`.

Can I reuse the same team across sessions?

Yes. All project state is persisted in a SQLite database. Any agent can rejoin a project by reading the current summary and picking up where the team left off. Projects support lifecycle transitions (active, paused, archived, closed).

How does AgentTeam avoid redundant work?

Specialists share structured research artifacts via the `share_artifact` tool, and the Project Manager orchestrates tasks so no agent re-researches what another has already found. The team protocol is lazy-loaded to save ~6,000 words of context.

Comments

More Other MCP servers