@promptibus/mcp

Model intelligence for AI agents — syntax, parameters, and routing for 67+ generative AI models, delivered via Model Context Protocol.

Promptibus MCP gives your AI agent structured knowledge about generative AI models: which model fits a task, how to format prompts for it, what parameters to use, and what pitfalls to avoid. It's not an API wrapper — it doesn't generate images or music. Instead, it tells your agent how to use the tools it already has access to.

Think of it as a prompt engineering co-pilot embedded in your agent's tool chain.

Quick Start

Add to your MCP client configuration (Claude Desktop, Cursor, Windsurf, etc.):

{
  "mcpServers": {
    "promptibus": {
      "command": "npx",
      "args": ["-y", "@promptibus/mcp"],
      "env": {
        "PROMPTIBUS_API_KEY": "psy_your_api_key_here"
      }
    }
  }
}

That's it. The package talks to the hosted Promptibus API — no database, no server setup.

| Variable | Required | Description | |---|---|---| | PROMPTIBUS_API_KEY | No | API key for higher rate limits and full tool access. Works anonymously without one. Get a key at promptibus.com/settings/api-keys | | PROMPTIBUS_API_URL | No | Override API base URL (default: https://promptibus.com). Useful for testing or self-hosted Promptibus instances. |

Tools

All seven tools are available to every tier — including anonymous use without an API key. Tiering applies to rate limits and which models you can query against (see below).

| Tool | Description | Example Input | |---|---|---| | recommend_model | Find the best model for a task. Returns top 3 with reasoning and parameters. | { "task": "photorealistic portrait", "domain": "IMAGE" } | | optimize_prompt | Optimize a prompt for a specific model. Applies model-specific syntax, community-tested parameters, and best-practice wording. | { "text": "a cat in space", "model": "midjourney-v7" } | | lint_prompt | Lint a prompt against a model's rules. Finds deprecated flags, invalid parameters, or length violations, and suggests fixes. | { "prompt": "a cat --ar 16:9", "model": "flux-2-pro" } | | compare_models | Side-by-side comparison of 2-5 models with provider, domain, cost, and capabilities. | { "models": ["flux-2-pro", "midjourney-v7"], "criteria": "photorealism" } | | get_parameters | Get recommended parameters for a model, including defaults and community-tested configs. | { "model": "stable-diffusion-3-5", "task_type": "portrait" } | | get_model_profile | Complete model profile: capabilities, syntax guide, parameters, community tips, and related prompts. | { "model": "suno-v4" } | | get_pricing | Real-world USD pricing for a model, a domain, or a planned volume. Includes cheaper alternatives and total-cost estimates. | { "model": "dall-e-3", "volume": 100 } |

Resources

Model profiles are available as MCP resources at:

promptibus://models/{slug}

Each resource returns a Markdown document with the model's provider, domain, version, pricing, description, and full guide content.

Prompts

The system-prompt prompt provides access to curated system prompts from the Promptibus community.

Without a slug: lists all available system prompts across IMAGE, VIDEO, AUDIO, TEXT, CODE domains
With a slug: returns the full prompt content, ready to use as a system message

# List all
system-prompt

# Get specific
system-prompt { "slug": "midjourney-prompt-architect" }

Authentication

Authentication is optional but recommended. Without an API key, you get anonymous-tier access.

Create an account at promptibus.com
Go to Settings > API Keys
Generate a key (starts with psy_)
Set PROMPTIBUS_API_KEY in your MCP config

Rate Limits

All tiers get access to all 6 tools. The difference is how many calls you get per day and which models you can query against.

| Plan | Daily Limit | Model coverage | |---|---|---| | Anonymous (no key) | 25 requests | 10 free-tier models | | Free (with key) | 100 requests | 10 free-tier models | | Pro | 500 requests | All 67+ models | | Studio | 2,000 requests | All 67+ models |

Limits reset daily at midnight UTC. See pricing for plan details.

Caching

To keep things fast and reduce unnecessary API traffic, the client caches responses for four tools whose output changes rarely: get_model_profile, get_parameters, compare_models, get_pricing. Cache TTL is 24 hours, in-memory (per process). Cache is skipped for tools whose output is input-dependent in a way that would get stale (recommend_model, optimize_prompt, lint_prompt).

Supported Models

67+ models across 5 domains:

AUDIO — Suno v4, v3.5 | Udio v1.5 | ElevenLabs | Stable Audio 2.0 | and more

CODE — Claude 4 Sonnet | GPT-4o | Gemini 2.5 Pro | DeepSeek V3 | and more

Full list at promptibus.com/models.

Privacy

Your prompts and tool calls are sent to promptibus.com over HTTPS
API keys are hashed server-side (SHA-256); the raw key never appears in logs
Anonymous usage (no key) is rate-limited by IP
No client-side DB, no local state beyond a single HTTP client

License

MIT

MCP Servers