OpenProject MCP Server

An MCP (Model Context Protocol) server that connects Claude Desktop to your OpenProject instance. Once set up, you can manage your projects, tasks, and time entries just by talking to Claude — no clicking around in the UI required.

What You Can Do

Just talk to Claude like you would a teammate. Here are some real examples:

Creating work

"Create a bug in the Pinnacle Power project called 'Login page throws 500 on mobile'. Assign it to Nathan, set it to High priority."

"Add a task under the TimeSheet - App feature for updating the punch-in endpoint. Set the module to match the feature."

Browsing and searching

"Show me all open work packages assigned to Gabe in the Fleet project."

"Search for anything related to 'penta integration' across all projects."

"What work packages are currently in beta testing?"

Updating work

"Mark work package 2583 as Passed QA."

"Set the percentage done on task 2791 to 75%."

"Reassign work package 2612 to Sarah and set it to Medium priority."

Commenting and logging time

"Add a comment to work package 2596 saying the API endpoint has been deployed to staging."

"Log 3.5 hours against work package 2598 for today."

Managing the cache

"Refresh the cache — I just added some new users and projects."

"Add a new feature to the cache: ID 2900, subject 'New Reporting Dashboard', project Pinnacle Power, module Dash - Reports."

Related MCP server: Procrastinator MCP Server

How the Cache Works

Most AI tools make a fresh API call every time they need basic info like "what projects exist?" or "what's Nathan's user ID?". That gets slow and noisy.

This server keeps a local SQLite database (dist/cache.db) on your machine with your projects, users, work package types, statuses, and open features baked in. When Claude needs to look something up, it reads from the cache instantly — no API round-trip needed.

The database lives in dist/ and is never committed to git (your credentials aren't in it either).

Cache tools Claude can use:

The features cache is special: if you have a custom field called "Module" (which we use to track various features which may be related yet aren't part of the same parent) and you set that Module field value in the feature, Claude will automatically set the Module field of the new work package to match that feature. This keeps your module assignments consistent without you having to remember them.

Setup

1. Prerequisites

• Node.js 18 or newer — download here

• An OpenProject instance (self-hosted or cloud)

• An OpenProject API key — in OpenProject, go to your avatar > My Account → Access Tokens and create one

2. Clone and Build

git clone https://github.com/your-username/openproject-mcp-server.git
cd openproject-mcp-server
npm install
npm run build

The build compiles the TypeScript and creates an empty dist/cache.json. You'll fill it in a moment.

3. Set Your Credentials

Copy the example env file:

cp .env.example .env

Then edit .env with your details:

OPENPROJECT_URL=https://your-openproject-instance.example.com
OPENPROJECT_API_KEY=your_api_key_here

The .env file is gitignored and never leaves your machine.

Connecting to Claude Desktop

Open your Claude Desktop config file in a text editor:

• Windows: %APPDATA%\Claude\claude_desktop_config.json

• macOS: ~/Library/Application Support/Claude/claude_desktop_config.json

Add the openproject entry inside "mcpServers":

{
  "mcpServers": {
    "openproject": {
      "command": "node",
      "args": ["/absolute/path/to/openproject-mcp-server/dist/index.js"],
      "env": {
        "OPENPROJECT_URL": "https://your-openproject-instance.example.com",
        "OPENPROJECT_API_KEY": "your_api_key_here"
      }
    }
  }
}

Replace the path with the actual location where you cloned the repo. On Windows it looks like:

C:/Users/yourname/source/repos/openproject-mcp-server/dist/index.js

Restart Claude Desktop. You should see the OpenProject tools listed in a new conversation (look for the hammer/tools icon).

First Run: Populate the Cache

The cache starts empty. Just ask Claude to fill it:

"Rebuild the cache from OpenProject."

Claude will call rebuild_cache, which fetches your projects, users, types, statuses, and open features from the live API and saves them to dist/cache.json. Takes a few seconds. You only need to do this once — or whenever things change significantly.

Keeping the Cache Fresh

The cache doesn't auto-update. It reflects whatever was true when you last rebuilt or edited it. Here's how to keep it current:

Full rebuild — when a lot has changed (new users onboarded, projects added, etc.):

"Rebuild the cache."

Targeted updates — for small, specific changes:

"Add user ID 60, login jsmith@example.com, name Jane Smith, status active to the cache."

"Remove project 19 from the cache."

"Update feature 2596 in the cache — it's now In Beta Testing."

You can also edit dist/cache.json directly in any text editor. It's a plain JSON file with five arrays: projects, users, types, statuses, and features.

Full Tool Reference

Cache (instant — no API call) get_cached_projects, get_cached_users, get_cached_types, get_cached_statuses, get_cached_features, rebuild_cache, cache_upsert_project, cache_delete_project, cache_upsert_user, cache_delete_user, cache_upsert_type, cache_delete_type, cache_upsert_status, cache_delete_status, cache_upsert_feature, cache_delete_feature

Projects list_projects, get_project

Work Packages list_work_packages, get_work_package, create_work_package, update_work_package, delete_work_package, search_work_packages

Activities & Comments list_work_package_activities, add_work_package_comment

Users list_users, get_user, get_current_user

Reference Data list_statuses, list_types, list_project_types, list_priorities, list_versions, list_categories

Time Tracking list_time_entries, create_time_entry

Relations & Attachments list_work_package_relations, list_work_package_attachments

Queries list_queries

Fork It, Copy It, Make It Yours

This is MIT licensed — do whatever you want with it. Fork it, copy pieces of it, adapt it for a different project management tool. No need to ask.

The structure here (API wrapper + local cache + CRUD tools exposed via MCP) maps pretty cleanly onto any REST API. If you're building something similar for Jira, Linear, GitHub Projects, or anything else, the patterns should give you a solid starting point.

If you build something cool on top of it, feel free to share — but no obligation.