Comfy MCP is an officially maintained, local-first MCP server from Comfy-Org built on comfy-cli. It runs as a stdio subprocess of your client and drives the ComfyUI on your own machine, covering generation, job monitoring, output collection, live introspection of your installed nodes/models/templates, and workflow validation and management. It is local-first but not local-only: partner models run on partner infrastructure, and COMFYUI_URL can target a machine you control. It has no path to Comfy Cloud, so tie into the separate remote HTTP server for cloud GPU execution. It needs Python >= 3.10 and comfy-cli >= 1.14.0, is beta, and exposes around 40 tools. It suits users with local inference hardware who want to automate a local ComfyUI from an AI client; credit spend, network exposure and local environment changes are the main risks, each behind a confirmation gate that still deserves careful authorization.
Strongest · Documentation 17/20 Weakest · Reliability 10/20
Static review · not runListed 2026-09-10
Read the FMRS scoring method →Runtime：Python 3.10+ · Prebuilt binary comfy-cli >= 1.14.0
COMFY_API_KEY
Comfy API key, only needed for partner-API nodes (Seedream/Veo/Kling etc.).
COMFY_BIN
Absolute path to the comfy-cli binary; needed when comfy is not on the PATH of the environment your client uses to launch the server.
COMFYUI_URL
Full http URL to a ComfyUI on another machine (e.g. http://gpu-box:8188); remote-targets the submit/job tools.
COMFYUI_HOST
Hostname of a remote ComfyUI, paired with COMFYUI_PORT (default 8188).
COMFYUI_PORT
Port of a remote ComfyUI; only valid together with COMFYUI_HOST.
COMFY_LOCAL_URL
Non-default address of a ComfyUI on this machine (read by comfy-cli), e.g. http://127.0.0.1:8189.
COMFY_MCP_REMOTE_SHARED_MODELS
Set to 1 when this machine's models dir is the remote's shared storage, skipping the download_model remote guard.
COMFY_MCP_ASSUME_CONSENT
Pre-authorize confirmation gates: install_node, update_all, version_switch, kill_untracked, network_exposure, or all.
COMFY_MCP_DEBUG_LOG
Enables the opt-in failure log; replaces the old COMFY_LOCAL_MCP_DEBUG_LOG.
COMFY_T2I_TEMPLATE
Retargets generate_image to a different local text-to-image template; set together with both SLOT vars.
COMFY_T2I_PROMPT_SLOT
Slot key of the positive prompt in the replacement template (e.g. 6.text); set with COMFY_T2I_TEMPLATE.
COMFY_T2I_CHECKPOINT_SLOT
Slot key of the checkpoint in the replacement template (e.g. ckpt_name); set with COMFY_T2I_TEMPLATE.
{
"mcpServers": {
"comfy-mcp": {
"command": "comfy-mcp",
"env": {
"COMFY_BIN": "/path/to/venv/bin/comfy",
"COMFY_API_KEY": "<your-comfy-api-key>"
}
}
}
}
Shown for Claude Code. Other clients may use a different file or key (VS Code uses "servers") — the configurator below converts it.
{
"servers": {
"comfy-mcp": {
"command": "comfy-mcp",
"env": {
"COMFY_BIN": "/path/to/venv/bin/comfy",
"COMFY_API_KEY": "<your-comfy-api-key>"
}
}
}
}
Goes in your project's .vscode/mcp.json (VS Code uses a "servers" key).
claude mcp add comfy-mcp -e COMFY_BIN=/path/to/venv/bin/comfy -e 'COMFY_API_KEY=<your-comfy-api-key>' -- comfy-mcp
Run it in a terminal; replace any <…> placeholders with your own values first.
After adding the server to your client config and restarting, the tool list should show server_info, run_workflow, fetch_outputs and the other tools; run `comfy-mcp --version` to confirm install, then ask the agent to call server_info to confirm the local ComfyUI is up.
Once connected, you can ask your AI assistant things like:
No matching tools
Listed from the project's documentation, not tested by this site.
Comfy MCP is a local MCP server maintained by Comfy-Org and built on comfy-cli. It is launched as a stdio subprocess by your AI client and drives the ComfyUI installed on your own machine (default 127.0.0.1:8188, or one on another machine you control via COMFYUI_URL / COMFYUI_HOST). It can generate images, submit and monitor jobs asynchronously, surface failure verdicts, collect output PNGs, introspect the nodes/models/templates your live install actually has, validate and edit workflows, and launch/stop/restart ComfyUI and tail its logs. Every tool shells out to the comfy command with --where local --json and parses comfy-cli's envelope/1 output. It requires Python >= 3.10, comfy-cli >= 1.14.0, a ComfyUI workspace and a running ComfyUI. It is currently beta, with 40 tools. Note it has no path to Comfy Cloud; for cloud-hosted execution use the separate remote HTTP server at https://cloud.comfy.org/mcp.
Manage your Hevy workout data from AI assistants via MCP.
Firecrawl's official MCP server for web search, scraping, and structured extraction for AI agents
One binary. One grammar. Evidence from the biomedical sources you already trust.
Encrypted, fully offline long-term memory for AI agents
Source revision 57098a2c8175 Data synced 2026-10-19 Read the FMRS scoring method