ClickUp MCP Server
Enables Claude Code and Claude Desktop to manage ClickUp tasks, sprints, reports, workspace navigation, and docs through 25+ MCP tools.
README
ClickUp MCP Server
A Model Context Protocol (MCP) server that connects Claude Code and Claude Desktop to ClickUp. Provides task management, sprint tracking, reporting, workspace navigation, and Doc creation through 25 tools.
Features
- Task management — create, update, search, and bulk-edit tasks
- Sprint tracking — auto-detect current sprint, list sprint tasks, filter by assignee/status
- Reporting — sprint reports with at-risk detection, PR link extraction, status summaries
- Workspace navigation — browse spaces, folders, lists; resolve custom task IDs
- Comments — read and write task comments
- Docs — create a native ClickUp Doc with real markdown tables/headers, unlike a task description
Prerequisites
- Python 3.12+
- uv (
brew install uv) - A ClickUp personal API token
Quick Start
uv tool install "clickup-mcp-server @ git+https://github.com/asaphe/clickup-mcp.git"
clickup-mcp-server setup --code
The setup script walks you through:
- Token retrieval (1Password integration or manual paste)
- Workspace configuration (workspace ID required, sprint/team IDs optional)
- Client registration (Claude Code, Claude Desktop, or both)
- Automatic restart of Claude Desktop if running
To configure Claude Desktop instead, run:
clickup-mcp-server setup --desktop
To configure both clients:
clickup-mcp-server setup --both
Alternative: Run from a Local Clone
Use this flow when you want to hack on the server itself:
git clone https://github.com/asaphe/clickup-mcp.git
cd clickup-mcp
uv sync --dev
uv tool install --editable .
python3 setup_mcp.py
This uses the same setup wizard, but runs it from the repo root layout used by this public package.
Run uv tool install --editable . first: the setup wizard only installs
clickup-mcp-server via uv tool install when it isn't already on your PATH,
so an editable install done beforehand is preserved instead of being replaced
by the published GitHub version.
Configuration
All configuration is via environment variables. Only CLICKUP_API_TOKEN and WORKSPACE_ID are required — the rest enable optional features.
Required
| Variable | Description |
|---|---|
CLICKUP_API_TOKEN |
Personal API token from https://app.clickup.com/settings/apps |
WORKSPACE_ID |
Your ClickUp workspace (team) ID |
Optional — Sprint Detection
| Variable | Description |
|---|---|
DEVELOPMENT_SPACE_ID |
Space ID containing your sprints. Required for sprint tools. |
SPRINTS_FOLDER_ID |
Folder ID within the space that holds sprint lists. Required for sprint tools. |
Optional — Team Labels
| Variable | Description |
|---|---|
COMPONENT_TEAM_FIELD_ID |
Custom field ID for Component/Team labels |
CLICKUP_TEAM_LABELS |
JSON mapping of team names to label IDs (see below) |
Example CLICKUP_TEAM_LABELS:
{"backend": "uuid-1", "frontend": "uuid-2", "example-team": "uuid-3"}
Optional — Tuning
| Variable | Default | Description |
|---|---|---|
API_BASE_URL |
https://api.clickup.com/api/v2 |
ClickUp API base URL |
REQUEST_TIMEOUT |
15.0 |
HTTP request timeout in seconds |
MAX_RETRIES |
3 |
Max retry attempts for failed/rate-limited requests |
Finding Your IDs
- Workspace ID: Settings → Workspaces → look at the URL or API response
- Space ID: Click on a Space → the ID is in the URL (
/s/{space_id}/...) - Folder ID: Click on a Folder → inspect the URL or use the
get_workspace_hierarchytool - Custom Field ID: Use the ClickUp API:
GET /list/{list_id}/field - Team Label IDs: Use the ClickUp API:
GET /list/{list_id}/field→ find the labels dropdown field → extract option IDs
Tools
28 tools across 6 categories:
Sprint Management
| Tool | Description |
|---|---|
get_current_sprint |
Auto-detect the active sprint by date |
refresh_sprint_cache |
Force-refresh the cached sprint |
get_sprint_tasks |
List tasks in the current sprint (filter by assignee/status) |
Task Management
| Tool | Description |
|---|---|
get_task |
Get task details by ID (DEV-1234 or UUID) |
create_task |
Create a task in any list |
create_sprint_task |
Create a task in the current sprint (auto-assign, team, points) |
update_task |
Update status, assignee, description, points, dates |
search_tasks |
Search tasks by name |
get_my_tasks |
Get tasks assigned to the current user |
get_list_tasks |
Get all tasks in a specific list |
move_task_to_list |
Move a task to a different list |
bulk_update_tasks |
Batch-update status/team/assignee across multiple tasks |
ensure_task_fields |
Check and fix missing fields (assignee, team, points) |
delete_task |
Delete a task permanently |
Comments
| Tool | Description |
|---|---|
add_task_comment |
Post a comment on a task |
get_task_comments |
Retrieve comments from a task |
Reporting
| Tool | Description |
|---|---|
get_sprint_report |
Sprint report by assignee with status counts and at-risk flags |
Workspace
| Tool | Description |
|---|---|
get_current_user |
Show the authenticated user |
get_workspace_hierarchy |
Browse spaces, folders, and lists |
task_url |
Get the ClickUp URL for a task |
add_tag_to_task |
Add a tag to a task |
list_teams |
List available team labels |
check_team_labels |
Check configured team labels against live ClickUp options |
Docs
| Tool | Description |
|---|---|
create_doc |
Create a native ClickUp Doc (v3 API) with one page of markdown content; returns its URL. Create-only, PRIVATE by default — sharing is a manual follow-up in the ClickUp UI. |
update_doc_page |
Overwrite an existing Doc page's content in place (replace-only), with an optional retitle; returns its URL. |
get_doc |
Get a Doc's metadata (name, parent location, visibility). |
get_doc_pages |
Get every page of a Doc, content included, as a flat list — sub-pages are flattened in, not dropped. |
get_doc_page |
Get a single page's content by doc_id/page_id. |
ClickUp's v3 Docs API has no DELETE endpoint for Docs or Pages — confirmed against
developer.clickup.com's own llms.txt index and public OpenAPI spec, and empirically
(a live DELETE call against a real doc returned 405 Method Not Allowed, not 404,
meaning the route exists but the verb is rejected). So there's no delete_doc tool, and
update_doc_page is replace-only by design: the client retries transport errors,
including a timeout that can fire after the server already applied the edit — a replace
is idempotent under retry, but append/prepend would silently double-apply.
ClickUp's public API also has no Doc/page comment endpoint (only task/list/view comments exist) — reading a Doc's comment panel still requires the browser.
Usage Examples
In Claude Code or Claude Desktop, just ask naturally:
- "show my tasks"
- "sprint report for backend"
- "create a task for fixing the login bug"
- "mark DEV-1234 as done"
- "what's the current sprint?"
- "show unassigned tasks in the sprint"
- "write this up as a doc so the tables render properly"
- "read that doc back to me"
Key Patterns
- Task IDs: Both custom IDs (
DEV-1234,PROJ-456) and UUIDs are accepted everywhere - "me" as assignee: Resolved automatically from the API token
- Sprint auto-detection: Current sprint is detected by date and cached per session
- Rate limiting: Automatic retry with exponential backoff on 429 responses
- Concurrent comment fetching: Sprint reports fetch PR links from comments with bounded concurrency
Coexistence with Built-in ClickUp MCP
This server works alongside the official ClickUp MCP connector. The built-in handles features not covered here (time tracking, chat, and a Doc's comment panel — ClickUp's public API has no Doc-comment endpoint). Doc create/read/update are covered here; deletion isn't possible via either server, since the v3 Docs API exposes no delete endpoint. You can register both, though having both may cause tool-selection ambiguity for overlapping operations.
Development
uv sync --dev
uv run pytest
uv run mypy clickup_mcp_server
uv run ruff check .
Architecture
clickup_mcp_server/
server.py — FastMCP server entry point and instructions
client.py — Async HTTP client with retry and rate-limit handling
config.py — Settings (env var based, no hardcoded IDs)
models.py — Pydantic models for API responses
tools/
sprint.py — Sprint detection and caching
tasks.py — Task CRUD, search, bulk operations
comments.py — Comment read/write
reporting.py — Sprint reports with at-risk detection
workspace.py — User info, hierarchy, tags
docs.py — Doc create/read/update (v3 API)
License
MIT
Recommended Servers
playwright-mcp
A Model Context Protocol server that enables LLMs to interact with web pages through structured accessibility snapshots without requiring vision models or screenshots.
Magic Component Platform (MCP)
An AI-powered tool that generates modern UI components from natural language descriptions, integrating with popular IDEs to streamline UI development workflow.
Audiense Insights MCP Server
Enables interaction with Audiense Insights accounts via the Model Context Protocol, facilitating the extraction and analysis of marketing insights and audience data including demographics, behavior, and influencer engagement.
VeyraX MCP
Single MCP tool to connect all your favorite tools: Gmail, Calendar and 40 more.
graphlit-mcp-server
The Model Context Protocol (MCP) Server enables integration between MCP clients and the Graphlit service. Ingest anything from Slack to Gmail to podcast feeds, in addition to web crawling, into a Graphlit project - and then retrieve relevant contents from the MCP client.
Kagi MCP Server
An MCP server that integrates Kagi search capabilities with Claude AI, enabling Claude to perform real-time web searches when answering questions that require up-to-date information.
E2B
Using MCP to run code via e2b.
Neon Database
MCP server for interacting with Neon Management API and databases
Exa Search
A Model Context Protocol (MCP) server lets AI assistants like Claude use the Exa AI Search API for web searches. This setup allows AI models to get real-time web information in a safe and controlled way.
Qdrant Server
This repository is an example of how to create a MCP server for Qdrant, a vector search engine.