Crewhu MCP Server

An MCP server for the Crewhu Open API, the customer-satisfaction and employee-recognition platform used by MSP service desks. It lets an AI agent read CSAT survey responses (including detractors and promoters), employees, badges and the recognition feed, the prize catalog and prize redemptions, and, on explicit request, post External Metric results to contest leaderboards.

• Transport: MCP Streamable HTTP at POST /mcp, health probe at GET /health

• Stateless and multi-tenant: the token arrives on every request as an HTTP header and is never stored; every tool is listed from the first tools/list

• Python 3.12, uv, FastMCP, httpx

• License: Apache-2.0, Copyright 2026 MSPbots.ai

Authentication headers

Where to get it:

• In the Crewhu web app open Settings -> Integrations -> API, click +Add, give the token a description, Confirm, then Copy Token.

• crewhu_badges_update_contest additionally needs Contests update via API enabled under Settings -> Contests, and a contest that uses an External Metric.

How the token is sent to Crewhu: every call goes to https://api.crewhu.com/api/v1/... with the token in the X_CREWHU_APITOKEN header (the name Crewhu expects, with underscores).

A request to /mcp without the header gets HTTP 401 naming it.

Related MCP server: hudu-mcp

Tools

All tools return a JSON string. List tools take limit (max 100, Crewhu's limit) and step (page number from 1) and return the records plus {count, total, step, has_more, next_step}. Ids are Crewhu _id strings; dates are ISO 8601 UTC. Survey records get a computed sentiment (positive for rating 5, neutral for 0, negative for -5).

Errors come back as a JSON envelope instead of an exception:

{"error": {"code": "unauthorized", "message": "Crewhu returned HTTP 401 for survey: API_ERR_INV_AUTH_PARAM (the API token is not valid)", "retryable": false}}

code is one of not_configured, unauthorized, not_found, invalid_argument, rate_limited, upstream_error; Crewhu's error tokens (API_ERR_*) are kept and explained. Reads are retried with backoff on 429, 5xx and network errors; the one write is retried only on 429.

Run locally

uv sync
uv run python -m crewhu_mcp          # listens on 0.0.0.0:8080

or with Docker:

docker build --platform linux/amd64 -t crewhu-mcp:dev .
docker run --rm -p 8080:8080 crewhu-mcp:dev

Configuration (environment variables, all optional): MCP_HTTP_HOST (default 0.0.0.0), MCP_HTTP_PORT (default 8080), LOG_LEVEL (default INFO). The token is never read from the environment.

Testing example

# list tools
curl -s http://localhost:8080/mcp \
  -H "Content-Type: application/json" \
  -H "Accept: application/json, text/event-stream" \
  -H "X-Crewhu-Api-Token: your-crewhu-api-token" \
  -d '{"jsonrpc":"2.0","id":1,"method":"tools/list","params":{}}'

# negative surveys since June
curl -s http://localhost:8080/mcp \
  -H "Content-Type: application/json" \
  -H "Accept: application/json, text/event-stream" \
  -H "X-Crewhu-Api-Token: your-crewhu-api-token" \
  -d '{"jsonrpc":"2.0","id":2,"method":"tools/call","params":{"name":"crewhu_surveys_detractors","arguments":{"since":"2026-06-01T00:00:00Z"}}}'

Unit tests (no network access needed):

uv run pytest

Delivery notes

• Tool count (18). Kept as is because the tool names are an existing contract.

• crewhu_badges_update_contest parameters changed. Older versions took id, target, name and description, which match no documented Crewhu endpoint. The tool now sends the documented POST /v1/contest/metrics body: metric_token, timeframe and data ([{identificator, value}], where identificator is an employee id or email). Values replace the leaderboard's current ones; it is the only write tool.

• Client-side filtering. The Open API filters only by update date, active flag and record fields, and has no text search. Sentiment, search and "pending" status are matched here: detractors / promoters read up to 1,000 surveys in the window, pending redemptions up to 1,000 redemptions, and the search tools the first limit records. Results say how many records were scanned and whether the scan was complete.

• JSON output. Results are structured JSON (with a computed survey sentiment) instead of formatted text.

• Single records. *_get tools read /v1/{resource}/{id} and fall back to an _id filter on the collection.

License

Apache License 2.0 - see LICENSE and NOTICE.