Live on npm — @promptibus/[email protected]
Your AI agent doesn't know
how generative models work.
It can write code and answer questions — but it can't tell you the right Midjourney V8 parameters, Suno V5 tag syntax, or Runway Gen-4 physics prompting. Promptibus MCP gives it that knowledge.
agentI need product photos on white background. Which model, what params?
mcp
recommend_model → Flux 2.0 Pro: API-native, $0.04/image, fast
get_parameters →
optimize_prompt → rewritten for Flux with negative prompts
get_parameters →
guidance_scale: 3.5, steps: 28, aspect_ratio: "1:1"optimize_prompt → rewritten for Flux with negative prompts
3 calls, 1 correct result. Without MCP: 3-5 guesses, wasted tokens.
LLMs are blind to generative models
ChatGPT, Claude, and Gemini know the world of text. They don't know the specific syntax, parameters, and constraints of the models that actually generate images, video, audio, and code.
Midjourney V7/V8
--oref, --ow, --sref, --cref, --personalize
Omni-reference, style reference, character reference — undocumented in LLM training data
Suno V5
[Verse], [Drop], [Falsetto], genre tags
Metatag processing that controls song structure and vocal style
Runway Gen-4
Physics-based prompting, camera motion syntax
Describing physical interactions and camera moves that Gen-4 actually understands
Stable Diffusion / ComfyUI
LoRA triggers, negative prompts, sampler configs
Model-specific keywords, weighting syntax like (word:1.3), and sampler parameters
The result: your agent guesses. 3-5 attempts instead of 1. Wasted API calls, wasted credits, bad outputs.
80
AI models tracked
5
domains
760+
curated prompts
Weekly
data updates
7 tools your agent can call
Each tool returns structured data with links back to promptibus.com for full model guides, community tips, and prompt galleries.
recommend_modelrecommend_model(task, constraints)
ask: "Product photos on white background, budget under $0.05/image"
get: Flux 2.0 Pro — API-native, $0.04/image, 28-step generation, built-in background removal
optimize_promptoptimize_prompt(text, model)
ask: "A cat floating in space" → Midjourney V8
get: cosmic feline drifting weightlessly through nebula, bioluminescent fur, volumetric starlight --ar 16:9 --s 350 --style raw --v 8
compare_modelscompare_models(models[], criteria)
ask: ["midjourney-v8", "flux-2-pro", "dall-e-3"] for "photorealism"
get: Side-by-side: resolution, pricing, API access, photorealism rating, strengths/weaknesses
get_parametersget_parameters(model, task_type)
Install in 30 seconds
Three install paths: remote HTTP (no install), npm package (offline-capable), or one-click through Smithery. All free-tier by default — no API key required.
One-click installRecommended
Smithery handles the config for Claude Desktop, Cursor, Windsurf, and other hosts — pick your client and click install.
Remote HTTP endpointZero install
Point any MCP client that supports HTTP transport at promptibus.com/api/mcp. No npm, no process management — works behind the firewall.
Any MCP client that supports HTTP transport
{
"mcpServers": {
"promptibus": {
"url": "https://promptibus.com/api/mcp"
}
}
}Source on GitHub · npm · Smithery — MIT licensed, no API key required for the anonymous tier.
Free tier is generous by design
We want every agent using Promptibus. The free tier covers most use cases. Pay only when you need higher limits or advanced tools.
10 models included on every free tier
Anonymous & Free plans · no API key needed for anonymousThese cover the most common image, video, audio, text, and code generation tasks. Pro and Studio plans expand coverage to all 80 models in the catalog.
Not a prompt library. Model intelligence.
Community-maintained
760+ prompts and model tips contributed by practitioners who use these models daily. Not scraped docs — real working knowledge.
Always current
Midjourney V8 shipped last month. Suno V5 changed tag syntax. LLM training data is months behind. Promptibus data updates weekly.
No competition
As of April 2026, no other MCP server provides model intelligence for generative AI. Search MCP, web MCP, code MCP — model MCP is ours.
Frequently asked questions
Is the MCP server free?
Yes. Every tier — including anonymous — gets all 7 tools. Anonymous gets 25 calls/day and access to 10 most popular models; free account unlocks 100 calls/day; Pro/Studio expand to all 80+ models. No credit card required on free tiers.
How fresh is the data?
Pro and Studio tiers get updates weekly — the same day we verify new model changes. Free tier updates monthly. Model profiles always reflect the latest verified information.
Which models are covered?
80 models across 5 domains: image (Midjourney, Flux, DALL-E, Stable Diffusion), video (Runway, Sora, Kling), audio (Suno, Udio, ElevenLabs), text (ChatGPT, Claude, Gemini), and code (Copilot, Cursor, Claude Code).
Can I contribute tips and corrections?
Yes. Every model guide on promptibus.com accepts community edits. Contributions from the web app feed directly into MCP responses — your tips help every agent using the server.
Do I need an API key?
Anonymous tier works without a key (25 calls/day, 10 models). Create a free account and generate an API key for 100 calls/day. Pro and Studio tiers raise limits and expand model coverage to all 80+ models.
How do I install it?
npm install @promptibus/mcp, or use npx directly from your MCP client config. Works with Claude Desktop, Claude Code, Cursor, n8n, and any other MCP-compatible client. Full config examples above.
Is the source open?
Yes — MIT licensed at github.com/promptibus/mcp. The server is a thin stdio shim over the promptibus.com HTTP API; all model intelligence lives server-side and updates continuously.
Give your agent the knowledge it's missing
Install @promptibus/mcp in 30 seconds. Anonymous tier works without an account — create one for higher limits.