Remote MCP · Streamable HTTP · OAuth or key
MCP Image Generator — one server for images, video and audio
Give Claude Code, Cursor, VS Code, Codex, OpenCode or any MCP client real image generation. One hosted server, one URL, many models — GPT Image 2.5, Nano Banana 2.1, Seedream 5.0, Recraft — plus edits, video, music, speech and 3D, billed from your normal Kenerate credits.
- Hosted, nothing to install
- OAuth sign-in or ken_ key
- 29 tools, many models
- Same credits as the app
Updated by the Kenerate AI team
On this page
Quick answer
What is an MCP image generator?
An MCP image generator is an image tool your AI agent calls through the Model Context Protocol. Most open-source ones wrap a single provider and need that provider's API key on your machine. Kenerate MCP is a hosted remote server: add `https://mcp.kenerateai.com/mcp` to your client (transport Streamable HTTP), sign in or paste a `ken_` key, and the agent gets `generate_image` plus 28 more tools — edits, upscaling, video, music, speech and 3D — on one account. Images start at 7.2 credits (GPT Image 2.5, 1K).
Key facts
- Server URL
- https://mcp.kenerateai.com/mcp
- Transport
- Streamable HTTP (remote)
- Auth
- MCP OAuth sign-in, or a ken_ API key
- Tools
- 29 — images, video, audio, 3D, planning
- Rate limit
- 8 calls a second per account
Gallery
Four images from one generate_image call each
An agent sent these prompts to the Kenerate MCP generate_image tool on GPT Image 2.5 Sunburst — medium quality, 1K, 4:5, one try each, 7.2 credits each. The tool returns a hosted URL, so an agent can drop the file straight into a repo, a page or the next tool call.
How to
How to use the mcp image generator
- Time needed
- Steps
- 4 steps
Step 1
Add the server
Claude Code: claude mcp add --transport http kenerate https://mcp.kenerateai.com/mcp. Cursor: add { "mcpServers": { "kenerate": { "url": "https://mcp.kenerateai.com/mcp" } } } to ~/.cursor/mcp.json. Codex: codex mcp add kenerate --url https://mcp.kenerateai.com/mcp. Other clients: a remote Streamable HTTP server with that URL.
Step 2
Sign in or add a key
Clients with MCP OAuth (Claude Code /mcp → Authenticate, Cursor “Needs login”, codex mcp login kenerate, opencode mcp auth kenerate) open a Kenerate page: sign in and click Allow. Clients without sign-in, like LM Studio, take a ken_ key from /app/mcp as an Authorization: Bearer header.
Step 3
Check the connection
Ask the agent to check your Kenerate balance. If get_balance answers, the 29 tools are live.
Step 4
Ask for an image
“Make a 16:9 hero image for the landing page and save it to public/hero.webp.” The agent picks a model, calls generate_image and gets back a hosted URL; the result is also saved in your Kenerate history.
See it in the app

Open Image Studio and type what you want to see — subject, style, lighting and mood.
Watch the tour · 1:23
Prompts, styles, references, edit, Arena & chat
Chapters
Tips
Tips for agents that generate images
What works when a coding agent drives the server.
Ask the agent to call estimate_cost before a batch and get_balance first — both are free, so it never spends blind.
Name the model in the request: GPT Image 2.5 for text and UI mock-ups, Nano Banana 2.1 to keep a product or character the same across shots, Seedream 5.0 for illustration, Recraft V4.1 for logos.
Results come back as hosted URLs. Have the agent download them into your assets folder, or pass the URL on to upscale_image, remove_background or generate_video.
For edits, the agent uploads your file with upload_media first, then runs generate_image with it as a reference — local paths are not accepted.
Put exact on-image text in quotes; GPT Image 2.5 rendered “JAZZ AFTER DARK” letter for letter.
Keep keys out of project config files that go into git — use the user-level config or sign in with OAuth instead.
Models
Image models the MCP server can call
What each model is good at. Open any one straight in the generator.

OpenAI
GPT Image 2.5 Sunburst
- Reference images
- Up to 16
- Output
- 1K–4K
Open GPT Image 2.5 SunburstBest for
Text, UI, photoreal — made this page
Google
Nano Banana 2.1

ByteDance
Seedream 5.0 Pro
- Reference images
- Up to 10
- Output
- 1K · 2K
Open Seedream 5.0 ProBest for
Illustration, cheap volume

- Free plan

FAQ
MCP Image Generator questions
10 questions
What is an MCP image generator?
An image tool exposed through the Model Context Protocol, the standard way AI agents call outside tools. Instead of you opening an image app, the agent — Claude Code, Cursor, Codex and so on — calls a tool like generate_image with a prompt and gets the picture back. Kenerate MCP is a hosted one: nothing to install, many models behind one URL.
Link to this answerHow do I add image generation to Claude Code?
Run claude mcp add --transport http kenerate https://mcp.kenerateai.com/mcp (add --scope user to use it in every project), start claude, type /mcp and pick kenerate → Authenticate. Sign in to Kenerate in the browser and click Allow. Then ask Claude Code for an image. For claude.ai and Claude Desktop, see our Claude AI image generator page.
Link to this answerWhich MCP clients does it work with?
Any client that supports remote MCP servers over Streamable HTTP. /app/mcp has setup steps, checked against each client's docs, for Claude, Claude Code, Claude Desktop, ChatGPT (developer mode), Cursor, VS Code, Codex, Gemini CLI, Antigravity, Windsurf, Kimi Code, Zed, Cline, Roo Code, Goose, LM Studio, Perplexity and Manus. OpenCode works with a "type": "remote" entry in opencode.json and opencode mcp auth kenerate.
Link to this answerDo I need an API key?
Not for clients that support MCP sign-in (OAuth) — Claude, ChatGPT, Cursor, VS Code, Claude Code, Codex, Gemini CLI and most others: you click Allow on a Kenerate page. Keys starting with ken_ are for clients without sign-in (LM Studio, Roo Code) and for scripts. A key is shown once, only a hash is stored, and you can keep up to 10 and rotate or delete them on /app/mcp.
Link to this answerIs there a free MCP image generator?
You can start free: a new Kenerate account gets starter credits, and Qwen Image 3.0 is on the free plan. Other models use credits at the same price as the app — GPT Image 2.5 Sunburst is 7.2 credits for a 1K image. There is no separate MCP fee. Video and 3D need a paid plan.
Link to this answerWhat's the best image generation MCP for Claude?
It depends on what you need. Single-provider open-source servers are free to run but need that provider's key and give you one model. A hosted multi-model server like Kenerate lets the agent pick per job — GPT Image 2.5 for text and UI, Nano Banana 2.1 for consistent products, Seedream 5.0 for illustration — and adds edits, upscaling and video without another setup.
Link to this answerCan the agent edit or upscale my own images?
Yes. The agent sends your file with upload_media (a public URL or base64), then runs generate_image with it as a reference to edit it, or a dedicated tool: upscale_image, remove_background, expand_image, change_camera_angle, apply_filter or split_layers.
Link to this answerCan it make video, music and 3D too?
Yes — the same server has generate_video, talking_avatar, motion_control, restyle_video, add_subtitles, dub_video, generate_music, generate_speech and generate_3d. Video and 3D need a paid plan, the same rule as in the app.
Link to this answerIs there a rate limit?
8 calls a second per account, across all keys. Long jobs like video run in the background: the agent calls wait_for_generation until the result is ready.
Link to this answerIs it an alternative to Higgsfield MCP?
Both are hosted remote MCP servers for images and video that need no provider API key. Kenerate lists all 29 tools and the credit price of every model, supports OAuth or a key, and documents setup for 18 clients on /app/mcp. Compare the two apps on our Higgsfield AI alternative page.
Link to this answerIn depth
MCP Image Generator, explained
What works, what to avoid, and how to pick a model.
On this page
Config for the common clients
Every client below signs in with OAuth when no key is set; swap in an Authorization: Bearer ken_… header to use a key instead. The full list, with one-click install links, is on /app/mcp.
- Claude Code — claude mcp add --transport http kenerate https://mcp.kenerateai.com/mcp, then /mcp → Authenticate.
- Cursor — ~/.cursor/mcp.json: { "mcpServers": { "kenerate": { "url": "https://mcp.kenerateai.com/mcp" } } }, then Cursor Settings → MCP → Needs login.
- VS Code (Copilot agent mode) — user mcp.json: { "servers": { "kenerate": { "type": "http", "url": "https://mcp.kenerateai.com/mcp" } } }, then Start.
- Codex CLI — codex mcp add kenerate --url https://mcp.kenerateai.com/mcp, then codex mcp login kenerate.
- Gemini CLI — gemini mcp add --transport http kenerate https://mcp.kenerateai.com/mcp, then /mcp auth kenerate.
- OpenCode — opencode.json: { "mcp": { "kenerate": { "type": "remote", "url": "https://mcp.kenerateai.com/mcp" } } }, then opencode mcp auth kenerate.
- LM Studio — mcp.json with the URL and an Authorization: Bearer ken_… header (LM Studio has no documented MCP sign-in), then chat with a model that supports tool calls.
Hosted server vs a self-hosted image MCP
Most image MCP servers in directories are small open-source wrappers: you run them locally, set one provider's API key in an environment variable and get that provider's model. That's fine for one model and full control.
A hosted server moves all of that off your machine. There is no process to keep running and no provider keys to manage; the agent sees many models through one list_models call, can check the price with estimate_cost, and every result is stored in your Kenerate history with a hosted URL.
The 29 tools
What an agent can call once the server is connected:
- Images: generate_image (create or edit), upscale_image, remove_background, expand_image, change_camera_angle, apply_filter, split_layers.
- Video (paid plan): generate_video, upscale_video, add_subtitles, dub_video, talking_avatar, motion_control, restyle_video, genjutsu_video.
- Audio: generate_music, generate_speech. 3D (paid plan): generate_3d.
- Free helpers: get_balance, list_models, tool_guide, estimate_cost, pick_model, pick_voice, suggest_workflow, upload_media, get_generation, wait_for_generation, list_generations.
Safety and your data
Calls through MCP go through the same moderation as the Kenerate app, whatever the client or prompt: content involving minors, sexual content of real people and illegal content are always refused.
Results are saved to your Kenerate history. Raw keys are never logged. Your AI client also sees your prompts and results under its own privacy terms.













