# Agent resources for Prefect, FastMCP, and Dagster Human-readable hub: https://www.prefect.io/agents Use the official source links below to check current instructions before installation. Ask permission before installing tools or changing configuration. Keep credentials out of chat and version control. Confirm the workspace or server before acting. Prefect skills: coming soon. No public repository or installation command is available on this page. ## Claude Code Guide: https://www.prefect.io/agents/claude-code Install the Prefect plugin in Claude Code to inspect flow runs and search Prefect documentation while you work on your code. Prerequisites: Claude Code with plugin support and a Prefect Cloud account. The hosted plugin does not require a local Prefect installation. ### Add the marketplace Enter this in a Claude Code session. Context: Claude Code ```text /plugin marketplace add prefecthq/prefect-mcp-server ``` ### Install Prefect Install the plugin, then follow the browser sign-in to choose the Prefect Cloud workspaces it can access. Context: Claude Code ```text /plugin install prefect ``` ### Prompt Help me install the official Prefect plugin in Claude Code. Read https://github.com/PrefectHQ/prefect-mcp-server for the current instructions. Check whether it is already installed before making changes. Explain the marketplace and plugin commands I need to enter, and guide me through Prefect Cloud sign-in without asking me to paste credentials into chat. After setup, list the workspaces you can access and ask which one to use. Do not run or modify workflows. Claude Code plugin instructions: https://github.com/PrefectHQ/prefect-mcp-server#claude-code-plugin Prefect MCP documentation: https://docs.prefect.io/v3/how-to-guides/ai/use-prefect-mcp-server ## Codex Guide: https://www.prefect.io/agents/codex Add the Prefect plugin to Codex for workspace diagnostics and Prefect documentation alongside your project files. Prerequisites: A Codex version with plugin support and a Prefect Cloud account. The commands below run in your terminal. ### Add the marketplace Register Prefect’s official plugin repository. Context: Terminal ```text codex plugin marketplace add prefecthq/prefect-mcp-server ``` ### Install the plugin Install Prefect and complete the Cloud sign-in when prompted. Context: Terminal ```text codex plugin add prefect@prefect ``` ### Prompt Set up the official Prefect plugin for Codex using https://github.com/PrefectHQ/prefect-mcp-server. Check my installed Codex version and existing plugins first. Explain the changes and ask before installing. Guide me through Prefect Cloud OAuth; do not ask for credentials in chat. Then list accessible workspaces and ask which one to use. Do not create, run, or modify workflows. If plugin commands are unavailable, explain the supported MCP alternative from the same documentation. Codex plugin instructions: https://github.com/PrefectHQ/prefect-mcp-server#codex-plugin Prefect MCP documentation: https://docs.prefect.io/v3/how-to-guides/ai/use-prefect-mcp-server ## Cursor Guide: https://www.prefect.io/agents/cursor Connect Cursor to Prefect’s MCP server to look up runs, logs, and deployments from your editor. Prerequisites: Cursor, uv installed and available on PATH, and a configured Prefect profile or environment variables for your target server. This setup uses local credentials, not the hosted plugin’s sign-in. ### Configure your Prefect connection Use the MCP setup guide to configure your Cloud workspace or self-hosted server locally. Keep API keys out of committed files. Prefect connection setup: https://docs.prefect.io/v3/how-to-guides/ai/use-prefect-mcp-server ### Add the MCP configuration Merge this entry into your project’s .cursor/mcp.json. Keep any other servers already configured there. Context: .cursor/mcp.json ```text { "mcpServers": { "prefect": { "command": "uvx", "args": ["--from", "prefect-mcp", "prefect-mcp-server"] } } } ``` ### Enable the server in Cursor Check Cursor’s MCP settings for the Prefect tools. If the server cannot start, confirm Cursor can find uvx and your Prefect configuration. ### Prompt Help me connect Cursor to Prefect using https://docs.prefect.io/v3/how-to-guides/ai/use-prefect-mcp-server. Check whether uvx is available and whether this project already has .cursor/mcp.json. Ask which Prefect Cloud workspace or self-hosted server I want to use. With my approval, merge the official prefect-mcp stdio configuration into the existing file without removing other servers. Keep credentials out of chat and version control. Verify the connection and report the active target without changing any workflows. Cursor configuration reference: https://github.com/PrefectHQ/prefect-mcp-server#other-mcp-clients Prefect MCP documentation: https://docs.prefect.io/v3/how-to-guides/ai/use-prefect-mcp-server ## VS Code / Copilot Guide: https://www.prefect.io/agents/vscode Connect GitHub Copilot in VS Code to Prefect’s local MCP server. This guide covers the editor, not Copilot CLI or GitHub’s cloud agent. Prerequisites: Enable Copilot agent mode and MCP access in VS Code. Install uv and configure your Prefect Cloud workspace or self-hosted server on the machine where the MCP process will run. This uses standard MCP, not a dedicated Prefect plugin. ### Configure your Prefect connection Follow Prefect’s MCP guide to set your local profile or environment variables. For remote workspaces and dev containers, uv and Prefect credentials must be available in that environment. Prefect connection setup: https://docs.prefect.io/v3/how-to-guides/ai/use-prefect-mcp-server ### Add the workspace configuration Merge this entry into .vscode/mcp.json without replacing existing servers. VS Code uses a servers key, not mcpServers. Keep credentials out of the file. Context: .vscode/mcp.json ```text { "servers": { "prefect": { "type": "stdio", "command": "uvx", "args": ["--from", "prefect-mcp", "prefect-mcp-server"] } } } ``` ### Start the server and check its tools Run MCP: List Servers from the Command Palette, select prefect, and start it. Review the trust prompt, then enable the Prefect tools in Copilot’s agent mode. Ask which Prefect target is configured and request its five most recent flow runs without changing anything. ### Prompt Help me connect GitHub Copilot in VS Code to Prefect. Read https://docs.prefect.io/v3/how-to-guides/ai/use-prefect-mcp-server and https://code.visualstudio.com/docs/agent-customization/mcp-servers first. This is the VS Code editor setup, not Copilot CLI or the cloud agent. Check for uvx and an existing .vscode/mcp.json, including whether this workspace runs remotely or in a container. Ask which Prefect Cloud workspace or self-hosted server I want. Explain the changes and ask before merging a prefect entry under servers with type stdio, command uvx, and args ["--from", "prefect-mcp", "prefect-mcp-server"]. Preserve other servers and keep credentials out of chat and version control. Help me start and approve the server, confirm the configured target, and read five recent flow runs without modifying anything. VS Code MCP configuration: https://code.visualstudio.com/docs/agent-customization/mcp-servers Prefect’s VS Code and Copilot example: https://github.com/PrefectHQ/prefect-mcp-server#other-mcp-clients Prefect MCP documentation: https://docs.prefect.io/v3/how-to-guides/ai/use-prefect-mcp-server ## OpenCode Guide: https://www.prefect.io/agents/opencode Add Prefect’s local MCP server to OpenCode using your existing Prefect connection. Prerequisites: Install OpenCode and uv, then configure a Prefect profile or environment variables for your Cloud workspace or self-hosted server. This uses OpenCode’s standard MCP configuration, not a dedicated Prefect plugin. ### Configure your Prefect connection Set up local authentication using Prefect’s MCP guide. The server inherits your active Prefect profile unless environment variables override it. Prefect connection setup: https://docs.prefect.io/v3/how-to-guides/ai/use-prefect-mcp-server ### Add Prefect to OpenCode Merge this entry into your project’s opencode.json or existing opencode.jsonc. Preserve other configuration. OpenCode uses an mcp key and a command array. Context: opencode.json ```text { "$schema": "https://opencode.ai/config.json", "mcp": { "prefect": { "type": "local", "command": ["uvx", "--from", "prefect-mcp", "prefect-mcp-server"], "enabled": true } } } ``` ### Check the connection Reopen OpenCode in the project and check the server list. Ask it to use the Prefect tools to confirm the configured target and read five recent flow runs without changing anything. Context: Terminal ```text opencode mcp list ``` ### Prompt Connect OpenCode to Prefect using https://docs.prefect.io/v3/how-to-guides/ai/use-prefect-mcp-server and https://opencode.ai/docs/mcp-servers/. Check for uvx and existing OpenCode configuration before editing. Ask which Prefect Cloud workspace or self-hosted server I want. With my approval, merge an mcp.prefect entry with type local, command ["uvx", "--from", "prefect-mcp", "prefect-mcp-server"], and enabled true into the appropriate existing config. Preserve other settings and keep credentials out of chat and version control. Check opencode mcp list, confirm the active Prefect target, and read five recent flow runs without changing workflows. OpenCode MCP configuration: https://opencode.ai/docs/mcp-servers/ Prefect MCP documentation: https://docs.prefect.io/v3/how-to-guides/ai/use-prefect-mcp-server ## Windsurf Guide: https://www.prefect.io/agents/windsurf Connect Windsurf’s Cascade agent to Prefect’s local MCP server to inspect runs and execution logs. Prerequisites: This configuration is for Cascade. Current Windsurf documentation redirects to Devin Desktop and distinguishes Cascade from Devin Local, which has a different configuration. Confirm which agent you use before editing. Install uv and configure local Prefect authentication; your team must allow MCP servers. ### Confirm your agent and Prefect target Check that you are using Cascade, then configure your Cloud workspace or self-hosted server using Prefect’s MCP guide. For Devin Local, follow the configuration linked from the current Cascade documentation instead of the file below. Current Cascade MCP documentation: https://docs.windsurf.com/windsurf/cascade/mcp ### Add the Cascade configuration Open Cascade’s MCP settings and edit the raw configuration. Merge the prefect entry into your existing file without removing other servers. Keep credentials in your local Prefect profile or secure environment, not in this file. Context: ~/.codeium/windsurf/mcp_config.json ```text { "mcpServers": { "prefect": { "command": "uvx", "args": ["--from", "prefect-mcp", "prefect-mcp-server"] } } } ``` ### Enable Prefect in Cascade Refresh the MCP server list and enable the Prefect tools. Ask Cascade to confirm the configured Prefect target and show five recent flow runs without changing anything. ### Prompt Help me connect Prefect to Windsurf’s Cascade agent. Read https://docs.prefect.io/v3/how-to-guides/ai/use-prefect-mcp-server and https://docs.windsurf.com/windsurf/cascade/mcp first. Confirm whether I am using Cascade or Devin Local: this setup targets Cascade; do not apply its file format to Devin Local. Check for uvx and existing MCP configuration. Ask which Prefect workspace or self-hosted server I want. With my approval, merge a prefect entry under mcpServers in Cascade’s existing configuration using command uvx and args ["--from", "prefect-mcp", "prefect-mcp-server"]. Preserve other servers, keep credentials out of chat and version control, and help me enable the tools. Confirm the target and read five recent flow runs without changing workflows. Windsurf / Cascade MCP configuration: https://docs.windsurf.com/windsurf/cascade/mcp Prefect’s Windsurf example: https://github.com/PrefectHQ/prefect-mcp-server#other-mcp-clients Prefect MCP documentation: https://docs.prefect.io/v3/how-to-guides/ai/use-prefect-mcp-server ## Gemini CLI Guide: https://www.prefect.io/agents/gemini-cli Add Prefect’s MCP tools to Gemini CLI using a local Prefect connection. Prerequisites: Gemini CLI, uv installed and available on PATH, and local Prefect authentication for your Cloud workspace or self-hosted server. ### Configure local authentication Set up your Prefect profile or environment variables before starting the MCP server. See the guide for Cloud and self-hosted options. Prefect connection setup: https://docs.prefect.io/v3/how-to-guides/ai/use-prefect-mcp-server ### Register the server Run this in your terminal, then check that Gemini CLI can discover the Prefect tools. Context: Terminal ```text gemini mcp add prefect uvx --from prefect-mcp prefect-mcp-server ``` ### Prompt Help me add Prefect’s MCP server to Gemini CLI. Use https://github.com/PrefectHQ/prefect-mcp-server for the supported configuration. Check for an existing Prefect MCP entry and for uvx before changing anything. Ask which Cloud workspace or self-hosted server I want to connect. Explain the changes, get my approval, and preserve existing servers. Keep credentials out of chat. Verify the active target and inspect recent runs without modifying workflows. Gemini CLI configuration reference: https://github.com/PrefectHQ/prefect-mcp-server#other-mcp-clients Prefect MCP documentation: https://docs.prefect.io/v3/how-to-guides/ai/use-prefect-mcp-server ## Claude Guide: https://www.prefect.io/agents/claude Connect your Prefect Cloud workspace to Claude to ask about workflow runs and their logs. Prerequisites: A Claude account with connector access and a Prefect Cloud account. No terminal setup is needed. ### Open the Prefect connector Add Prefect from Claude’s connector directory. Open Prefect in Claude: https://claude.ai/directory/prefect ### Sign in to Prefect Cloud Complete the sign-in and select the workspaces Claude can access. Approve tool calls when Claude asks. ### Prompt List the Prefect workspaces you can access. Ask me which workspace to use, then show its five most recent flow runs with their states. Do not change anything. Prefect MCP documentation: https://docs.prefect.io/v3/how-to-guides/ai/use-prefect-mcp-server ## ChatGPT Guide: https://www.prefect.io/agents/chatgpt Use Prefect in ChatGPT to investigate workflow runs and check your workspace. Prerequisites: ChatGPT with access to the Prefect plugin and a Prefect Cloud account. No local Prefect installation is needed. ### Add the Prefect plugin Open the directory listing and add Prefect to ChatGPT. Open Prefect in ChatGPT: https://chatgpt.com/plugins/plugin_asdk_app_6a5fe5e3471881919daf83971db15505?q=prefect ### Connect your account Start a chat with Prefect selected, then follow the Connect prompt to sign in to Prefect Cloud. Adding the plugin alone does not connect your account. ### Prompt List the Prefect workspaces you can access. Ask me which workspace to use, then show its five most recent flow runs with their states. Do not change anything. Prefect MCP documentation: https://docs.prefect.io/v3/how-to-guides/ai/use-prefect-mcp-server ## Other MCP clients Guide: https://www.prefect.io/agents/other-mcp-clients Configure Prefect’s MCP server in another agent or editor. The connection uses standard MCP; available tools and setup details depend on your client. Prerequisites: Check your client’s documentation for local stdio MCP support. Install uv and configure the Prefect Cloud workspace or self-hosted server you want to access. MCP compatibility alone does not mean every client or feature has been tested with Prefect. ### Configure local Prefect authentication Use your active Prefect profile or environment variables to select a Cloud workspace or self-hosted API. Keep credentials out of chat and committed configuration files. Prefect connection setup: https://docs.prefect.io/v3/how-to-guides/ai/use-prefect-mcp-server ### Register a local stdio server In your client’s MCP settings, name the server prefect, choose stdio, set the command to uvx, and supply the three arguments separately: --from, prefect-mcp, prefect-mcp-server. The equivalent command is shown below; let your client start the process. JSON formats differ, so use your client’s own schema. Context: Server launch command ```text uvx --from prefect-mcp prefect-mcp-server ``` ### Verify the target and tools Start the connection and review any trust or tool permissions. Ask your agent to confirm the configured Prefect target and read five recent flow runs without modifying anything. If your client only supports remote HTTP connections, use the deployment instructions in Prefect’s MCP guide instead of this local command. Remote MCP deployment instructions: https://docs.prefect.io/v3/how-to-guides/ai/use-prefect-mcp-server#cloud-deployment ### Prompt Help me connect my MCP client to Prefect using https://docs.prefect.io/v3/how-to-guides/ai/use-prefect-mcp-server. First ask which client and version I use, and which Prefect Cloud workspace or self-hosted server I want. Check the client’s official documentation for supported transports and its configuration schema; do not assume another client’s JSON format works. For local stdio support, use command uvx with arguments ["--from", "prefect-mcp", "prefect-mcp-server"]. If only remote HTTP is supported, explain Prefect’s documented deployment alternative and stop for my approval before deploying anything. Check existing configuration, explain proposed changes, and ask before installing or editing. Preserve other servers and keep credentials out of chat and version control. Verify the target and read recent runs without changing workflows. Prefect MCP documentation: https://docs.prefect.io/v3/how-to-guides/ai/use-prefect-mcp-server Prefect MCP source: https://github.com/PrefectHQ/prefect-mcp-server ## Prefect Guide: https://www.prefect.io/agents/prefect Inspect runs through MCP, use the CLI from your project, and look up current Prefect documentation. ### Connect your workspace Choose an agent guide for plugin or MCP setup. Hosted connections use Prefect Cloud sign-in; local MCP servers use your own Prefect configuration. Choose an agent: https://www.prefect.io/agents#connect ### Use the CLI for project work A coding agent can use the Prefect CLI in your development environment. Unlike the read-only MCP tools, CLI commands can change your workspace. Review commands that deploy, run, or delete resources. CLI setup for agents: https://www.prefect.io/agents/prefect-cli ### Prefect skills The public skills repository is coming soon. Installation instructions will be added here when it is available. ### Prompt Use Prefect to investigate the most recent failed flow run in my chosen workspace. Read its task runs and logs, cite the run and relevant errors, and explain what I should check before retrying. Do not retry or change anything. MCP setup and authentication: https://docs.prefect.io/v3/how-to-guides/ai/use-prefect-mcp-server Plugins and MCP source: https://github.com/PrefectHQ/prefect-mcp-server Install Prefect: https://docs.prefect.io/v3/get-started/install ## FastMCP Guide: https://www.prefect.io/agents/fastmcp Give your coding agent the FastMCP documentation, then use it to build and test MCP tools in Python. ### Add FastMCP to your project In an existing uv project, add the Python package. Check the installation guide for requirements and other package managers. Context: Terminal ```text uv add fastmcp ``` FastMCP documentation: https://gofastmcp.com ### Connect the documentation MCP server Add this URL in your agent’s MCP settings. This server searches FastMCP documentation; it does not connect the agent to your application’s data. Context: Documentation MCP URL ```text https://gofastmcp.com/mcp ``` ### Prompt Help me set up FastMCP in this project. Read https://gofastmcp.com/llms.txt and the current installation instructions. Check the Python environment, package manager, and existing dependencies first. Ask before installing anything or changing configuration. Help me connect the documentation MCP server at https://gofastmcp.com/mcp using my agent’s supported settings. Verify the installation locally without publishing a server or connecting production credentials. FastMCP documentation: https://gofastmcp.com Documentation index for agents: https://gofastmcp.com/llms.txt FastMCP on GitHub: https://github.com/PrefectHQ/fastmcp ## Dagster Guide: https://www.prefect.io/agents/dagster Install Dagster skills for development or connect an agent to your Dagster+ deployment through MCP. ### Install Dagster skills Follow Dagster’s skill installation instructions for your agent. The public repository contains guidance for working on Dagster projects. Dagster skill installation: https://docs.dagster.io/getting-started/ai-tools/skills ### Connect to Dagster+ The MCP server requires Dagster+ and is currently in beta. Choose the US or EU endpoint in the documentation and authenticate with OAuth. Its tools can change deployment state using your permissions; review actions before approving them. Dagster+ MCP setup: https://docs.dagster.io/getting-started/ai-tools/dagster-mcp ### Prompt Help me set up an agent for my Dagster project. Read https://docs.dagster.io/getting-started/ai-tools/skills and https://docs.dagster.io/getting-started/ai-tools/dagster-mcp. Check the project and existing agent configuration first. Ask which agent I use and whether I need local development skills, a Dagster+ connection, or both. For Dagster+, ask my region and guide me through sign-in without requesting secrets in chat. Get approval before installation. Verify access without launching runs or changing assets. Dagster skills repository: https://github.com/dagster-io/skills Dagster+ MCP documentation: https://docs.dagster.io/getting-started/ai-tools/dagster-mcp ## Set up the Prefect CLI for your agent Guide: https://www.prefect.io/agents/prefect-cli Install Prefect in your project and verify the environment your coding agent will use. ### Install in your project For an existing uv project, run the commands below. Use the installation guide if you use a different package manager. Do not install a second copy into a project that already has Prefect. Context: Terminal · uv project ```text uv add prefect uv run prefect version ``` Prefect installation guide: https://docs.prefect.io/v3/get-started/install ### Choose your server For Prefect Cloud, run the login command and complete browser authentication. For a self-hosted server, configure its API URL and authentication instead of running Cloud login. Context: Terminal · Prefect Cloud ```text uv run prefect cloud login ``` Cloud and self-hosted configuration: https://docs.prefect.io/v3/how-to-guides/ai/use-prefect-mcp-server ### Review commands before execution The CLI has write capabilities. Confirm the target workspace and review deployment, run, and deletion commands before allowing an agent to execute them. ### Prompt Set up the Prefect CLI for this project using https://docs.prefect.io/v3/get-started/install. Inspect the existing Python environment, package manager, and Prefect installation first. Propose any dependency changes and ask before installing. Verify the CLI version in the same environment you will use. Ask whether I use Prefect Cloud or a self-hosted server, then guide me through the appropriate authentication without exposing secrets. Confirm the target workspace or API URL. Do not deploy, start runs, or modify remote resources. Prefect installation documentation: https://docs.prefect.io/v3/get-started/install