upservice-mcp

upservice-mcp

Manages employees, projects, sprints, tasks, tags, directories, and channel messages via the Upservice Public API.

Category
Visit Server

README

Upservice MCP Server

Читать на русском

An MCP (Model Context Protocol) server that wraps the Upservice Public API so Claude (or any MCP client) can manage employees, projects, sprints, tasks, tags, directories, and channel messages directly.

Built from the OpenAPI spec at https://public.upservice.io/openapi.json, verified against the live API.

What it can do

58 tools covering:

  • Employees — list employees
  • Projects — list, create, get, update, delete, set managers/members, mark completed
  • Sprints — list, create, get, update, delete, complete, activate, add tasks
  • Tags — list, create, update, delete, assign/unassign to tasks/chats/assets/contacts/attachments
  • Tasks — list/search, create, get, update, delete, attachments, status changes, effort estimation, worklog, agreement (approval) workflow actions, co-responsibles, agreement/acquaintance sheets
  • Directories (custom reference catalogs) — list/create/get/update/delete directories and their records, plus bulk relation management (linking records to tasks, projects, orders, etc.)
  • Channels & files — load/send chat messages, send channel messages, upload files, get file download URLs

Every write tool exposes the well-documented fields explicitly (validated), and most also accept an extra_fields dict for advanced/uncommon fields documented in the Upservice API but not modeled individually.

Mentioning employees: Upservice only turns @Name into a real, notifying mention if it's written as @[Name](employee_id). Tools that write text (upservice_send_chat_message, upservice_send_channel_message, upservice_create_task, upservice_update_task) accept a mentions: [{employee_id, display_name}] field — put a {{employee_id}} placeholder in your text and the server substitutes the correct syntax for you.

Known API limitation: GET /v1/tasks (used by upservice_list_tasks) has no status/is_completed filter — confirmed against the live API, not just the docs. Filter by status client-side after narrowing with date_end_gte/date_end_lte, project, author, or responsible.

Setup

Get your Upservice API key first: in your Upservice account, go to account settings → API key. Keys look like UPS-XXXX-XXXX-XXXX-XXXX.

Option A — uvx (recommended, no manual install)

Requires uv installed (brew install uv or see their docs). Add to your MCP client config (e.g. claude_desktop_config.json):

{
  "mcpServers": {
    "upservice": {
      "command": "uvx",
      "args": ["--from", "/absolute/path/to/upservice_mcp", "upservice-mcp"],
      "env": {
        "UPSERVICE_API_KEY": "UPS-XXXX-XXXX-XXXX-XXXX"
      }
    }
  }
}

If this project is published to a git remote, --from can point at it directly instead of a local path: "args": ["--from", "git+https://github.com/alexherbaly/upservice-mcp", "upservice-mcp"] — then a colleague only needs the URL and their own API key, no file copying at all.

Pin to a ref rather than the bare URL — without one, uvx tracks the main branch, so a bad push to main breaks everyone's server on their next run. Two options:

  • @stable (recommended for colleagues) — a tag that CI automatically moves to the tip of main every time the build passes. No one has to remember to cut a release for routine fixes; everyone's server just picks up the latest known-good commit next time it restarts.
  • @v0.2.0 (a specific version) — frozen forever at that exact commit, for when you want a name you can point back to later (release notes, "the version we tested on date X"). Cut these manually with git tag when it's worth a name, not on every push.

Example: "args": ["--from", "git+https://github.com/alexherbaly/upservice-mcp@stable", "upservice-mcp"]

uv builds and caches an isolated environment on first run; nothing is installed system-wide.

Option B — plain venv + pip

cd upservice_mcp
python3 -m venv venv
venv/bin/pip install -r requirements.txt
{
  "mcpServers": {
    "upservice": {
      "command": "/absolute/path/to/upservice_mcp/venv/bin/python3",
      "args": ["/absolute/path/to/upservice_mcp/src/upservice_mcp/server.py"],
      "env": {
        "UPSERVICE_API_KEY": "UPS-XXXX-XXXX-XXXX-XXXX"
      }
    }
  }
}

Restart Claude Desktop afterwards. The tools will appear prefixed with upservice_.

Optional: custom base URL

If Upservice ever changes the API host, or you use a private/on-prem instance, override it:

"env": {
  "UPSERVICE_API_KEY": "UPS-XXXX-XXXX-XXXX-XXXX",
  "UPSERVICE_API_BASE_URL": "https://public.upservice.io"
}

Testing locally

# uvx
UPSERVICE_API_KEY=UPS-XXXX uvx --from . upservice-mcp

# venv
UPSERVICE_API_KEY=UPS-XXXX venv/bin/python3 src/upservice_mcp/server.py

Or test interactively with the MCP Inspector:

npx @modelcontextprotocol/inspector uvx --from . upservice-mcp

Notes

  • Authentication uses a raw API key in the Authorization header (not a Bearer token) — this server handles that for you.
  • Destructive tools (delete_*, unassign_tag) are annotated with destructiveHint: true so MCP clients can warn users appropriately.
  • Some request bodies documented by Upservice have many optional fields; where a tool doesn't model a field explicitly, pass it via extra_fields (a plain JSON object) and it will be merged into the request body.
  • Never share your own UPSERVICE_API_KEY — each person should generate their own from their Upservice account settings.

Recommended Servers

playwright-mcp

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.

Official
Featured
TypeScript
Magic Component Platform (MCP)

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.

Official
Featured
Local
TypeScript
Audiense Insights MCP Server

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.

Official
Featured
Local
TypeScript
VeyraX MCP

VeyraX MCP

Single MCP tool to connect all your favorite tools: Gmail, Calendar and 40 more.

Official
Featured
Local
graphlit-mcp-server

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.

Official
Featured
TypeScript
Kagi MCP Server

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.

Official
Featured
Python
Neon Database

Neon Database

MCP server for interacting with Neon Management API and databases

Official
Featured
E2B

E2B

Using MCP to run code via e2b.

Official
Featured
Exa Search

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.

Official
Featured
Qdrant Server

Qdrant Server

This repository is an example of how to create a MCP server for Qdrant, a vector search engine.

Official
Featured