GUIDE
Updated March 2026
8 topics

MCP Integration Guide

Learn how to connect Workunit to your development tools using the Model Context Protocol (MCP)

Overview

The Model Context Protocol (MCP) allows AI development tools to access Workunit directly. Once connected, AI assistants can create workunits, manage tasks, organize projects, link assets, and track your development work—all without leaving your editor or chat interface.

This guide covers installation for supported tools and provides a complete reference for all available MCP tools.

Fast setup

Most supported tools only need one install command, then a browser-based OAuth approval.

Shared context

Every connected AI client sees the same workunits, tasks, projects, and saved context.

OAuth secured

Access is authorized in your browser and persists across sessions for each tool.

Prerequisites

  • A Workunit account

    Sign up at workunit.app before you connect an MCP client.

  • An MCP-compatible AI tool

    Claude Code, OpenCode, Codex, Gemini CLI, or any other client that supports MCP over HTTP.

  • Internet access to workunit.app

    Your client needs to reach the hosted MCP endpoint and complete the OAuth flow in the browser.

Installation

Choose your AI development tool below for specific installation instructions.

Claude Code

Claude Code setup

Run this command in your terminal, then restart Claude Code if it is already open.

claude mcp add --transport http workunit https://workunit.app/mcp

OpenCode

OpenCode setup

Add the remote MCP server, then authorize it in your browser.

opencode mcp add
Location

Global

Server URL

https://workunit.app/mcp

Server Type

Remote over HTTP

Authentication

OAuth 2.0, no pre-registered client ID required

Then authenticate:

opencode mcp auth workunit
WSL2 Users

If the OAuth callback fails because the browser can not reach 127.0.0.1 inside WSL, copy the full callback URL from the browser and run it manually in your WSL terminal.

curl "http://127.0.0.1:19876/mcp/oauth/callback?code=...&state=..."

Codex (OpenAI)

Codex (OpenAI) setup

Add the Workunit MCP server, then approve the browser authorization prompt when it appears.

codex mcp add workunit --url https://workunit.app/mcp

Gemini CLI

Gemini CLI setup

Install the server first, then authenticate from inside Gemini CLI.

gemini mcp add --transport http workunit https://workunit.app/mcp

After installation, authenticate with:

/mcp auth workunit

Other MCP Tools

Other MCP Tools setup

For any MCP-compatible client, use these connection settings and follow that tool's standard remote-server setup flow.

Server URL

https://workunit.app/mcp

Transport

HTTP (Streamable)

Auth

OAuth 2.0

Authentication

Workunit uses OAuth 2.0 for secure authentication. The first time your AI tool uses a Workunit tool, you'll be prompted to authorize access through your browser.

1
Tool requests access

Your MCP client asks Workunit for permission the first time it needs a protected tool.

2
Browser opens

A browser window launches to the Workunit OAuth authorization screen.

3
Approve access

Review the requested permissions and click Authorize.

4
Return to your terminal

The tool is now connected and can call Workunit over MCP.

Your authorization persists across sessions. You only need to authenticate once per tool.

Try It Out

Now that your AI tool is connected, try these prompts to see Workunit in action. These work with any MCP-compatible client.

Verify your connection
What Workunit tools do you have access to?
Create your first workunit
I need to implement user authentication for my app.
Create a workunit for this with clear success criteria,
then break it down into tasks.
Plan work under a project
I'd like to build a REST API for my blog.
Create a project called "Blog API" and add workunits
for each endpoint I'll need.
Pick up where another AI left off
Search my workunits for anything related to "authentication".
Show me what's been done and what tasks are still open.
Implement an existing workunit
Get the workunit "Add password reset flow" and implement
the next open task. Update the task status when done.
Why this matters

Any AI model connected via MCP sees the same workunits, tasks, and context. Start planning in Claude, implement in Codex, review in Gemini — your progress follows you everywhere.

Available Tools

Once connected, your AI assistant has access to the following tools. These tools work identically across all MCP-compatible clients.

System

PUBLICping

Test MCP server connectivity

AUTHget_authenticated_user

Get details about the authenticated user

Projects

Projects organize related workunits and assets into logical groups.

AUTHcreate_project

Create a project to organize workunits and assets

AUTHget_project

Get project details with optional assets, checkins, and workunits

AUTHlist_projects

List projects with optional status/owner/tag filters and pagination

AUTHupdate_project

Update project fields via update_mask

AUTHasset_link

Link or unlink assets to/from projects or workunits. Specify project_id or workunit_id. Supports relationship_type.

Workunits

Workunits are focused pieces of work with problem statements and success criteria.

AUTHcreate_workunit

Create a workunit with problem statement and success criteria. Can link initial assets

AUTHget_workunit

Get workunit details with optional tasks, assets, and context atoms

AUTHupdate_workunit

Update workunit fields. Status=completed/archived triggers special workflows

Tasks

Tasks break workunits into actionable items with status tracking and dependencies.

AUTHcreate_task

Create a task in a workunit with title, priority, and optional dependencies

AUTHget_task

Retrieve task details including status, assignee, dependencies, comments, and time logs

AUTHupdate_task

Update task fields including status, priority, assignment, and due date

Assets

Assets represent products, people, knowledge, and systems that your work involves.

AUTHcreate_asset

Create assets of any type: product, people, knowledge, or system

AUTHget_asset

Get asset details by ID with type-specific fields

AUTHupdate_asset

Update asset fields via update_mask. Requires asset_type specification

Directories

Organize knowledge assets into a hierarchical folder structure.

AUTHdirectory

Manage directories for organizing knowledge assets. Actions: create, get, list, update, delete, move

Context

Save and retrieve structured context atoms — the trail-of-thought for a workunit.

AUTHsave_context

Save a context atom (decision, insight, question, attempt, or progress) to a workunit

Delete

Unified delete tool for permanently removing workunits, projects, or assets.

AUTHdelete

Permanently delete an entity by ID. Auto-detects type (workunit, project, or asset). Irreversible.

Troubleshooting

Connection failed

Verify your internet connection and ensure https://workunit.app is reachable. If you use a proxy or VPN, confirm it allows connections to the Workunit domain.

Tools not available

Restart your AI tool after installation. If tools still do not appear, remove and re-add the MCP server, then authenticate again.

Authentication issues

Complete the OAuth flow in your browser. If it fails, check whether third-party cookies are blocked for workunit.app or retry in an incognito window.

Tool errors

Most tool errors include a useful message. Common causes are missing required fields, invalid IDs, or insufficient permissions in the current organization.

claude mcp remove workunit && claude mcp add --transport http workunit https://workunit.app/mcp