openproject-mcp

openproject-mcp

MCP server for OpenProject that enables AI agents to search and view work packages, add comments, upload attachments, log time, list projects, and search users via the REST API v3.

Category
访问服务器

README

openproject-mcp

MCP (Model Context Protocol) server for OpenProject — gives AI agents (ZCode, Claude Desktop, etc.) a limited set of operations over OpenProject via the REST API v3:

  • 🔍 Search work packages — by subject, ID, status, project, assignee
  • 📄 Work package details — all fields, description, optional comments and attachments
  • 💬 Add comments to work packages
  • 📎 Upload files (attachments) to work packages
  • ⏱ Log time (time entries)
  • 📊 Time report for a period — grouped by project with a total
  • 🗂 List projects — all projects, parent's subprojects, or the full hierarchy tree
  • 👤 Search users — by name or login (to obtain IDs)

Русская версия

Three transports are supported (selected by the MCP_TRANSPORT variable or the --transport flag):

Transport Purpose
stdio Local clients (ZCode, etc.): the server runs as a subprocess and talks over stdin/stdout. Default.
streamable-http Modern MCP HTTP transport (endpoint /mcp). Recommended for Docker / a networked microservice.
sse Legacy HTTP transport (/sse + /messages/). For old clients that do not support streamable-http.

Why a custom server when OpenProject 17.2 has a built-in MCP? The built-in one is Enterprise-only and read-only. This server works with any edition (including Community) and supports write operations: comments, files, time.


Requirements

  • Python 3.10+ (tested on 3.13)
  • Access to an OpenProject instance with the API enabled (Personal Access Token)
  • Token permissions: view work packages/projects, add work package notes (comments), log time, add attachments (edit work package or add attachments)
  • For HTTP/Docker — Docker (or any ASGI server; uvicorn is included as a dependency)

Installation

Option A — via uv (recommended, faster)

cd path\to\openproject-mcp
uv venv
uv pip install -e ".[http]"     # [http] is only needed for the HTTP transport

Option B — via standard pip

cd path\to\openproject-mcp
python -m venv .venv
.venv\Scripts\activate
pip install -e ".[http]"        # for stdio, `pip install -e .` is enough

After installation both the openproject-mcp command and python -m openproject_mcp are available.

Configuration

Variables are grouped by prefix to avoid confusion:

  • op_ — connection to OpenProject (where we talk to)
  • mcp_ — settings of the MCP service itself (how it works)

Copy the example and fill in your values:

copy .env.example .env          # Windows
cp .env.example .env            # Linux/macOS

Environment variables

Variable Prefix Required Description
OP_URL op ✅ Base URL of your OpenProject without a trailing /. Example: https://openproject.example.com
OP_API_KEY op ✅ Personal API token. Created in profile settings → Access tokens. Requires the administrator setting "Enable API tokens".
MCP_TRANSPORT mcp ❌ stdio (default) | streamable-http | sse
MCP_BIND mcp ❌ HTTP transport address as host:port (IPv6: [address]:port). Default 127.0.0.1:8000. In Docker — 0.0.0.0:8000. Ignored for stdio.
MCP_AUTH_TOKEN mcp ❌ Optional Bearer token protecting the HTTP endpoint. Empty = no auth (trusted network / reverse proxy only). Clients send Authorization: Bearer <value>.
MCP_ALLOWED_HOSTS mcp ❌ Comma-separated host list (DNS-rebinding protection). Suffix :* — any port. Example: mcp.corp.local,mcp.corp.local:*. Empty = host protection disabled (see the "Host allowlist" section).
MCP_LOG_LEVEL mcp ❌ DEBUG / INFO / WARNING / ERROR. Default INFO. Logs go to stderr.

Variables can also be set without a .env — directly in the client config or at container startup.

Getting an API token

  1. Sign in to OpenProject.
  2. Profile icon (top right) → My account → Access tokens.
  3. Click + API token, give it a name (e.g. "MCP"), copy the value.
  4. If the section is unavailable, an administrator must enable Enable API tokens in Administration → … (or grant your account the right).

The token is shown only once — save it right away.


Running

stdio (local client)

openproject-mcp                 # MCP_TRANSPORT=stdio (default)

The server starts and waits for client commands over stdin/stdout. Stop with Ctrl+C.

streamable-http / sse (HTTP service)

openproject-mcp --transport streamable-http --bind 0.0.0.0:8000
# or via environment variables:
#   MCP_TRANSPORT=streamable-http MCP_BIND=0.0.0.0:8000 openproject-mcp

Health check:

curl http://127.0.0.1:8000/health        # → {"status": "ok"}  (no auth required)

CLI arguments (override env; precedence: CLI > env > default):

Argument Description
--transport {stdio,streamable-http,sse} Transport
--bind HOST:PORT Address for HTTP (IPv6: [address]:port). Ignored for stdio.
--log-level LEVEL DEBUG / INFO / WARNING / ERROR

Docker

The microservice is built into a portable image and runs as an HTTP service (streamable-http by default). Secrets (OP_URL, OP_API_KEY, MCP_AUTH_TOKEN) are passed at runtime — not baked into the image.

⚠️ Security. The server opens write operations to OpenProject (comments, files, time). On any network except a fully isolated one, set MCP_AUTH_TOKEN or keep the service behind an authenticated reverse proxy.

Build and run

# Build the image
docker build -t openproject-mcp .

# Run (secrets via -e / --env-file)
docker run --rm -p 8000:8000 \
  -e OP_URL=https://openproject.example.com \
  -e OP_API_KEY=your_api_token_here \
  -e MCP_AUTH_TOKEN=choose_a_secret \
  openproject-mcp

Health check:

curl http://localhost:8000/health                                    # 200
curl -H "Authorization: Bearer choose_a_secret" \
     -H "Content-Type: application/json" \
     -H "Accept: application/json, text/event-stream" \
     -X POST http://localhost:8000/mcp \
     -d '{"jsonrpc":"2.0","id":1,"method":"initialize","params":{"protocolVersion":"2025-11-25","capabilities":{},"clientInfo":{"name":"t","version":"1"}}}'

docker compose

Easier via docker-compose.yml (reads .env):

cp .env.example .env           # fill in OP_URL / OP_API_KEY / MCP_AUTH_TOKEN
docker compose up --build      # build + start
docker compose logs -f         # logs
docker compose down            # stop

After startup the MCP endpoint is http://localhost:8000/mcp, health at /health.

Container variables

The image sets the defaults MCP_TRANSPORT=streamable-http and MCP_BIND=0.0.0.0:8000 (overridable at runtime). The following are passed in explicitly:

  • OP_URL, OP_API_KEY — connection to OpenProject;
  • MCP_AUTH_TOKEN — endpoint protection (recommended);
  • MCP_ALLOWED_HOSTS — see below.

Host allowlist (DNS-rebinding protection)

By default the MCP SDK accepts HTTP requests only to localhost — in Docker (even on 0.0.0.0) or behind a reverse proxy this yields HTTP 421 on every request. This server behaves in a hybrid way:

  • If MCP_ALLOWED_HOSTS is set (e.g. mcp.corp.local,mcp.corp.local:*) — protection is enabled with that host list.
  • If empty — protection is disabled; the server relies on Bearer auth (MCP_AUTH_TOKEN), a reverse proxy, or network isolation.

If you see 421 "Invalid Host header" — either set MCP_ALLOWED_HOSTS with the hostname your clients connect to, or (for an internal network) leave it empty and protect the endpoint with MCP_AUTH_TOKEN.


Client integration

stdio (ZCode, Claude Desktop — local subprocess)

Add the configuration to your MCP client's settings file.

{
  "mcpServers": {
    "openproject": {
      "command": "C:\\path\\to\\project\\.venv\\Scripts\\python.exe",
      "args": ["-m", "openproject_mcp"],
      "env": {
        "OP_URL": "https://openproject.example.com",
        "OP_API_KEY": "your_api_token_here",
        "MCP_TRANSPORT": "stdio",
        "MCP_LOG_LEVEL": "INFO"
      }
    }
  }
}

Paths in JSON on Windows require a double backslash \\ or forward slashes. If a .env exists in the working directory, the env block can be omitted, but explicit variables are more reliable (they do not depend on the working directory at launch).

Alternative via console script (if openproject-mcp is on your PATH):

{
  "mcpServers": {
    "openproject": {
      "command": "C:\\path\\to\\project\\.venv\\Scripts\\openproject-mcp.exe",
      "args": [],
      "env": { "OP_URL": "https://openproject.example.com", "OP_API_KEY": "..." }
    }
  }
}

After saving the config, restart the client (or reconnect the MCP server). The tools with the op_* prefix will appear in the tool list.

streamable-http (remote microservice)

The client connects to the HTTP endpoint by URL and (if MCP_AUTH_TOKEN is set) sends the authorization header. The exact format depends on the client; for ZCode this is an MCP server section of type http/url:

{
  "mcpServers": {
    "openproject": {
      "type": "http",
      "url": "http://mcp.corp.local:8000/mcp",
      "headers": {
        "Authorization": "Bearer choose_a_secret"
      }
    }
  }
}

With MCP_TRANSPORT=sse the endpoints change to /sse (GET, stream) and /messages/ (POST) — use your client's SSE mode.


Tools

Tool Purpose Key parameters
op_check_connection Check URL + token —
op_search_work_packages Search work packages subject (a string or a list of synonyms, searched in subject with an automatic description fallback), status (open/closed/all), project_id, assignee_id, type_id, filters (JSON), page_size, offset, sort_by
op_get_work_package Details of one work package work_package_id, include_comments, include_attachments
op_list_projects List projects / subprojects / tree name (a string or a list of synonyms, with a description fallback), parent_id, direct_children_only, active, filters (JSON), page_size, offset, sort_by, as_tree
op_add_comment Comment on a work package work_package_id, comment, internal
op_add_attachment Upload a file work_package_id, file_path (absolute path), file_name
op_log_time Log time work_package_id, hours (1.5h/2h30m/90m/1:30/PT1H30M), activity_id, spent_on (YYYY-MM-DD), comment
op_list_time_entries Time report for a period user_id ('me' or ID), date_from/date_to (YYYY-MM-DD), project_id, activity_id, include_comments, page_size, offset, sort_by
op_list_users Search users query (name or login), filters (JSON), page_size, offset
op_list_time_entry_activities Time-entry activity reference —

Tools are available over any transport — behavior is identical for stdio and HTTP.

Usage examples

Find open work packages in project #5 assigned to me:

op_search_work_packages(project_id="5", status="open", assignee_id="me", page_size=10)

Find a work package by subject (one term or synonyms):

op_search_work_packages(subject="login")                         # one term
op_search_work_packages(subject=["bug", "defect", "issue"])      # synonyms → OR, deduplicated

Search goes through the work package subject first; if nothing is found it automatically falls back to the description. With no matches an empty list is returned (not the whole backlog).

Find a project by name:

op_list_projects(name="demo")
op_list_projects(name=["demo", "test"])          # synonyms, description fallback

All projects or the hierarchy tree:

op_list_projects()                              # flat list of all projects
op_list_projects(as_tree=True)                  # tree: roots → children → ...
op_list_projects(active=True)                   # only active projects

Subprojects of a specific project:

op_list_projects(parent_id="1")                              # full subtree (any depth)
op_list_projects(parent_id="1", direct_children_only=True)   # only direct children

Add a comment to work package #42:

op_add_comment(work_package_id=42, comment="Verified, the bug reproduces", internal=True)

Log 1.5 hours against work package #42:

op_log_time(work_package_id=42, hours="1.5h", activity_id=1, comment="Debugging")

Upload a file to work package #42:

op_add_attachment(work_package_id=42, file_path="C:\\reports\\bug.png")

Time report for a period (for me, for July):

# "for July" → date_from/date_to (the agent computes month boundaries itself)
op_list_time_entries(user_id="me", date_from="2026-07-01", date_to="2026-07-31")
# numbers only, no comments:
op_list_time_entries(user_id="me", date_from="2026-07-01", date_to="2026-07-31", include_comments=False)
# for a specific project:
op_list_time_entries(user_id="me", project_id="1", date_from="2026-07-01", date_to="2026-07-31")

Returns entries grouped by project, with per-project hour totals and a grand total.

Find a user by name (to substitute the ID):

op_list_users(query="Ivanov")
# then use the found id in op_list_time_entries(user_id="...")

OpenProject version compatibility

The server is not tied to a version number and works with any OpenProject exposing API v3. The only dialect-dependent point is the work-package link in a time entry:

  • OpenProject 14+ → _links.entity (/api/v3/work_packages/{id})
  • OpenProject ≤13 → _links.workPackage

op_log_time automatically tries the modern entity field and, on a server rejection (HTTP 422), retries with the legacy workPackage field. The successful variant is cached, so subsequent writes avoid extra attempts. Search, comments and attachments are identical across versions.


Project structure

openproject-mcp/
├── Dockerfile                 # microservice image (python:3.13-slim)
├── docker-compose.yml         # local compose startup
├── .dockerignore
├── pyproject.toml             # hatchling; deps: mcp[cli], httpx, anyio; extra [http]: uvicorn[standard]
├── .env.example               # config template (op_* / mcp_*)
├── .env                       # real credentials (in .gitignore)
└── src/openproject_mcp/
    ├── __init__.py            # package version
    ├── __main__.py            # entry point: transport selection (stdio / http), CLI, stderr logging
    ├── config.py              # .env / environment variable loading, validation, security_settings()
    ├── client.py              # httpx client for API v3: auth, HAL errors, pagination
    ├── formatting.py          # HAL+JSON _links parsing, ISO8601 durations, filters
    ├── http_app.py            # HTTP app assembly: /health + optional Bearer auth
    └── server.py              # MCP tool registration (transport-independent)

Troubleshooting

  • "Configuration error: OP_URL is not set" — no .env in the working directory and the variables were not passed via env/-e.
  • HTTP 401 Unauthorized — missing/incorrect MCP_AUTH_TOKEN. The client must send Authorization: Bearer <value>.
  • HTTP 421 "Invalid Host header" — the host allowlist triggered. Either set MCP_ALLOWED_HOSTS with the hostname clients use, or leave it empty (protection is disabled) and secure with MCP_AUTH_TOKEN.
  • Connection fails in Docker — check that MCP_BIND=0.0.0.0:8000 (not 127.0.0.1) and the port is published (-p 8000:8000).
  • "Port already in use" — change the port in MCP_BIND/--bind and in the port mapping.
  • HTTP 401/403 from OpenProject — wrong/expired token or missing permissions. Check the token and its rights (add work package notes, log time).
  • HTTP 404 — the work package/project was not found or you have no view permission.
  • Need diagnostics — set MCP_LOG_LEVEL=DEBUG; logs go to stderr.

Testing

pip install -r requirements-dev.txt
pytest tests/

The integration tests hit a live OpenProject server configured via OP_URL / OP_API_KEY (or the repo's .env) and are skipped automatically when the server is not reachable.

License

MIT

推荐服务器

Baidu Map

Baidu Map

百度地图核心API现已全面兼容MCP协议,是国内首家兼容MCP协议的地图服务商。

官方
精选
JavaScript
Playwright MCP Server

Playwright MCP Server

一个模型上下文协议服务器,它使大型语言模型能够通过结构化的可访问性快照与网页进行交互,而无需视觉模型或屏幕截图。

官方
精选
TypeScript
Magic Component Platform (MCP)

Magic Component Platform (MCP)

一个由人工智能驱动的工具,可以从自然语言描述生成现代化的用户界面组件,并与流行的集成开发环境(IDE)集成,从而简化用户界面开发流程。

官方
精选
本地
TypeScript
Audiense Insights MCP Server

Audiense Insights MCP Server

通过模型上下文协议启用与 Audiense Insights 账户的交互,从而促进营销洞察和受众数据的提取和分析,包括人口统计信息、行为和影响者互动。

官方
精选
本地
TypeScript
VeyraX

VeyraX

一个单一的 MCP 工具,连接你所有喜爱的工具:Gmail、日历以及其他 40 多个工具。

官方
精选
本地
graphlit-mcp-server

graphlit-mcp-server

模型上下文协议 (MCP) 服务器实现了 MCP 客户端与 Graphlit 服务之间的集成。 除了网络爬取之外,还可以将任何内容(从 Slack 到 Gmail 再到播客订阅源)导入到 Graphlit 项目中,然后从 MCP 客户端检索相关内容。

官方
精选
TypeScript
Kagi MCP Server

Kagi MCP Server

一个 MCP 服务器,集成了 Kagi 搜索功能和 Claude AI,使 Claude 能够在回答需要最新信息的问题时执行实时网络搜索。

官方
精选
Python
e2b-mcp-server

e2b-mcp-server

使用 MCP 通过 e2b 运行代码。

官方
精选
Neon MCP Server

Neon MCP Server

用于与 Neon 管理 API 和数据库交互的 MCP 服务器

官方
精选
Exa MCP Server

Exa MCP Server

模型上下文协议(MCP)服务器允许像 Claude 这样的 AI 助手使用 Exa AI 搜索 API 进行网络搜索。这种设置允许 AI 模型以安全和受控的方式获取实时的网络信息。

官方
精选