Summary

Reads broadly, writes narrowly.

Fourteen of the sixteen tools only read; the two that change anything are cancelling a run and starting one from a deployment, which is a sensible line for something an assistant drives. List and search calls default to a limit of 20, so ask explicitly when you want more.

What it is

An MCP server over the Prefect API that exposes flows, flow runs and deployments as tools. It covers the read side (get, list, search, filter) and the two write actions that matter day to day: cancelling a run and creating one from a deployment.

What you get

Flows fetched by ID or name, listed with pagination, searched by name and tags — `get_flow_by_id`, `get_flow_by_name`, `list_flows`, `search_flows`
Flow runs fetched by ID, listed with an optional flow filter, and searched by state such as COMPLETED, FAILED or CRASHED — `get_flow_run_by_id`, `list_flow_runs`, `search_flow_runs_by_state`
A running flow cancelled — `cancel_flow_run`
A new run created from a deployment with your own parameters and an optional wait — `create_flow_run_from_deployment`
Deployments fetched by ID or by `flow_name/deployment_name`, listed, and searched by whether their schedule is active — `get_deployment_by_id`, `get_deployment_by_name`, `list_deployments`, `search_deployments_by_status`
Raw Prefect filter criteria passed straight through when the named searches are not enough — `filter_flows`, `filter_flow_runs`, `filter_deployments`

Requirements

A reachable Prefect API. Set `PREFECT_API_URL` — the default is `http://localhost:4200/api` — and `PREFECT_API_KEY` if you are authenticating against a Prefect server or Prefect Cloud. Python 3.9 or newer and Prefect 3; install with `uv pip install -U prefect-mcp-server` and run it through `uvx prefect-mcp-server`. The package is `prefect-mcp-server` 1.1.3, MIT licensed, and has supported Prefect 3.0 since 1.1.2.

Setup effort

One command — uv pip install -U prefect-mcp-server