comfyui-mcp-server

Name: comfyui-mcp-server
Author: joenorton

390stars78forks

How to install comfyui-mcp-server?

joenorton/comfyui-mcp-server

$git clone https://github.com/joenorton/comfyui-mcp-server

Installs into the current project.

Summary

> Generate and refine AI images/audio/video through natural conversation

Files · 1

View on GitHub

README.md

1	# ComfyUI MCP Server
2
3	> Generate and refine AI images/audio/video through natural conversation
4
5	A lightweight MCP (Model Context Protocol) server that lets AI agents generate and iteratively refine images, audio, and video using a local ComfyUI instance.
6
7	You run the server, connect a client, and issue tool calls. Everything else is optional depth.
8
9	---
10
11	## Quick Start (2–3 minutes)
12
13	This proves everything is working.
14
15	### 1) Clone and set up
16
17	```bash
18	git clone https://github.com/joenorton/comfyui-mcp-server.git
19	cd comfyui-mcp-server
20	pip install -r requirements.txt
21	```
22
23	### 2) Start ComfyUI
24
25	Make sure ComfyUI is installed and running locally.
26
27	```bash
28	cd <ComfyUI_dir>
29	python main.py --port 8188
30	```
31
32	### 3) Run the MCP server
33
34	From the repository directory:
35
36	```bash
37	python server.py
38	```
39
40	The server listens at:
41
42	```
43	http://127.0.0.1:9000/mcp
44	```
45
46	### 4) Verify it works (no AI client required)
47
48	Run the included test client:
49
50	```bash
51	# Use default prompt
52	python test_client.py
53
54	# Or provide your own prompt
55	python test_client.py -p "a beautiful sunset over mountains"
56	python test_client.py --prompt "a cat on a mat"
57	```
58
59	`test_client.py` will:
60
61	* connect to the MCP server
62	* list available tools
63	* fetch and display server defaults (width, height, steps, model, etc.)
64	* run `generate_image` with your prompt (or a default)
65	* automatically use server defaults for all other parameters
66	* print the resulting asset information
67
68	If this step succeeds, the system is working.
69
70	Note: The test client respects server defaults configured via config files, environment variables, or `set_defaults` calls. Only the `prompt` parameter is required; all other parameters use server defaults automatically.
71
72	That’s it.
73
74	---
75
76	## Use with an AI Agent (Cursor / Claude / n8n)
77
78	Once the server is running, you can connect it to an AI client.
79
80	Create a project-scoped `.mcp.json` file:
81
82	```json
83	{
84	"mcpServers": {
85	"comfyui-mcp-server": {
86	"type": "streamable-http",
87	"url": "http://127.0.0.1:9000/mcp"
88	}
89	}
90	}
91	```
92
93	Note: Some clients use `"type": "http"` instead of `"streamable-http"`. Both work with this server. If auto-discovery doesn't work, try changing the type field.
94
95	Restart your AI client. You can now call tools such as:
96
97	* `generate_image`
98	* `view_image`
99	* `regenerate`
100	* `get_job`
101	* `list_assets`
102
103	This is the primary intended usage mode.
104
105	---
106
107	## What You Can Do After It Works
108
109	Once you’ve confirmed the server runs and a client can connect, the system supports:
110
111	* Iterative refinement via `regenerate` (no re-prompting)
112	* Explicit asset identity for reliable follow-ups
113	* Job polling and cancellation for long-running generations
114	* Optional image injection into the AI’s context (`view_image`)
115	* Auto-discovered ComfyUI workflows with parameter exposure
116	* Configurable defaults to avoid repeating common settings
117
118	Everything below builds on the same basic loop you just tested.
119
120	## Migration Notes (Previous Versions)
121
122	If you’ve used earlier versions of this project, a few things have changed.
123
124	### What’s the Same
125	- You still run a local MCP server that delegates execution to ComfyUI
126	- Workflows are still JSON files placed in the `workflows/` directory
127	- Image generation behavior is unchanged at its core
128
129	### What’s New
130	- Streamable HTTP transport replaces the older WebSocket-based approach
131	- Explicit job management (`get_job`, `get_queue_status`, `cancel_job`)
132	- Asset identity instead of ad-hoc URLs (stable across hostname changes)
133	- Iteration support via `regenerate` (replay with parameter overrides)
134	- Optional visual feedback for agents via `view_image`
135	- Configurable defaults to avoid repeating common parameters
136
137	### What Changed Conceptually
138	Earlier versions were a thin request/response bridge.
139	The current version is built around iteration and stateful control loops.
140
141	You can still generate an image with a single call, but you now have the option to:
142	- refer back to specific outputs
143	- refine results without

People who liked this also viewed

6 picks

Type

elevenlabs-mcpThe official ElevenLabs MCP serverMCP ServersJul 202653k1.5k
MonetEdit Videos and Design Images with Claude code or CodexMCP ServersJul 202646k97
scribe.jsJavaScript OCR and text extraction for images and PDFs.MCP ServersJul 202623k305
gliana-aiPay-per-call generative AI — 60+ models (image, video, music, speech). No signup, no API key; pay per call from your wallet over MPP/x402.MCP Serversrecently15k0
nanobanana-mcp-serverAI image generation MCP server powered by Google Gemini, with smart model selection and 4K outputMCP ServersMay 202612k381
FramesailCreate long-form (faceless YouTube) videos end to end from any MCP client: script, locked character references, storyboard, voiceover, and final video editing — with characters and style held consistent across every shot. Making long-form MCP Serversrecently11k0

comfyui-mcp-server

390stars78forks

View on GitHub

How to install comfyui-mcp-server?

joenorton/comfyui-mcp-server

$git clone https://github.com/joenorton/comfyui-mcp-server

Installs into the current project.

Summary

> Generate and refine AI images/audio/video through natural conversation

Files · 1

View on GitHub

README.md

1	# ComfyUI MCP Server
2
3	> Generate and refine AI images/audio/video through natural conversation
4
5	A lightweight MCP (Model Context Protocol) server that lets AI agents generate and iteratively refine images, audio, and video using a local ComfyUI instance.
6
7	You run the server, connect a client, and issue tool calls. Everything else is optional depth.
8
9	---
10
11	## Quick Start (2–3 minutes)
12
13	This proves everything is working.
14
15	### 1) Clone and set up
16
17	```bash
18	git clone https://github.com/joenorton/comfyui-mcp-server.git
19	cd comfyui-mcp-server
20	pip install -r requirements.txt
21	```
22
23	### 2) Start ComfyUI
24
25	Make sure ComfyUI is installed and running locally.
26
27	```bash
28	cd <ComfyUI_dir>
29	python main.py --port 8188
30	```
31
32	### 3) Run the MCP server
33
34	From the repository directory:
35
36	```bash
37	python server.py
38	```
39
40	The server listens at:
41
42	```
43	http://127.0.0.1:9000/mcp
44	```
45
46	### 4) Verify it works (no AI client required)
47
48	Run the included test client:
49
50	```bash
51	# Use default prompt
52	python test_client.py
53
54	# Or provide your own prompt
55	python test_client.py -p "a beautiful sunset over mountains"
56	python test_client.py --prompt "a cat on a mat"
57	```
58
59	`test_client.py` will:
60
61	* connect to the MCP server
62	* list available tools
63	* fetch and display server defaults (width, height, steps, model, etc.)
64	* run `generate_image` with your prompt (or a default)
65	* automatically use server defaults for all other parameters
66	* print the resulting asset information
67
68	If this step succeeds, the system is working.
69
70	Note: The test client respects server defaults configured via config files, environment variables, or `set_defaults` calls. Only the `prompt` parameter is required; all other parameters use server defaults automatically.
71
72	That’s it.
73
74	---
75
76	## Use with an AI Agent (Cursor / Claude / n8n)
77
78	Once the server is running, you can connect it to an AI client.
79
80	Create a project-scoped `.mcp.json` file:
81
82	```json
83	{
84	"mcpServers": {
85	"comfyui-mcp-server": {
86	"type": "streamable-http",
87	"url": "http://127.0.0.1:9000/mcp"
88	}
89	}
90	}
91	```
92
93	Note: Some clients use `"type": "http"` instead of `"streamable-http"`. Both work with this server. If auto-discovery doesn't work, try changing the type field.
94
95	Restart your AI client. You can now call tools such as:
96
97	* `generate_image`
98	* `view_image`
99	* `regenerate`
100	* `get_job`
101	* `list_assets`
102
103	This is the primary intended usage mode.
104
105	---
106
107	## What You Can Do After It Works
108
109	Once you’ve confirmed the server runs and a client can connect, the system supports:
110
111	* Iterative refinement via `regenerate` (no re-prompting)
112	* Explicit asset identity for reliable follow-ups
113	* Job polling and cancellation for long-running generations
114	* Optional image injection into the AI’s context (`view_image`)
115	* Auto-discovered ComfyUI workflows with parameter exposure
116	* Configurable defaults to avoid repeating common settings
117
118	Everything below builds on the same basic loop you just tested.
119
120	## Migration Notes (Previous Versions)
121
122	If you’ve used earlier versions of this project, a few things have changed.
123
124	### What’s the Same
125	- You still run a local MCP server that delegates execution to ComfyUI
126	- Workflows are still JSON files placed in the `workflows/` directory
127	- Image generation behavior is unchanged at its core
128
129	### What’s New
130	- Streamable HTTP transport replaces the older WebSocket-based approach
131	- Explicit job management (`get_job`, `get_queue_status`, `cancel_job`)
132	- Asset identity instead of ad-hoc URLs (stable across hostname changes)
133	- Iteration support via `regenerate` (replay with parameter overrides)
134	- Optional visual feedback for agents via `view_image`
135	- Configurable defaults to avoid repeating common parameters
136
137	### What Changed Conceptually
138	Earlier versions were a thin request/response bridge.
139	The current version is built around iteration and stateful control loops.
140
141	You can still generate an image with a single call, but you now have the option to:
142	- refer back to specific outputs
143	- refine results without

People who liked this also viewed

6 picks

Type