Connect your agent to every model it needs.
Give this to your agent: muapi.ai/SKILL.md and let it take it from there. Image, video, audio, lipsync, 3D, LLMs, and SEO/data tools — one API key, one skill file, no hardcoded model list, provider names hidden throughout.
657+ models, one API
Browse the catalog →The workflow
discover()
List or search the live model catalog — no hardcoded model names.
GET https://api.muapi.ai/api/v1/models?category=Text%20to%20Image # CLI muapi discover "cinematic image to video" --output-json
inspect()
Get a model's full input schema, required fields, and pricing.
GET https://api.muapi.ai/api/v1/models/{name}
# CLI
muapi inspect kling-3 --output-jsonestimate()
Preview the exact cost of a payload before spending any credits.
POST https://api.muapi.ai/api/v1/models/{name}/estimate-cost
# CLI
muapi estimate kling-3 --input payload.jsonrun()
Submit the payload to the model's real endpoint (resolved in step 2).
POST https://api.muapi.ai{endpoint}
-> {"request_id": "...", "status": "processing"}
# CLI
muapi run kling-3 --input payload.json --no-waitpoll()
Check status until the run completes, then read the output URLs.
GET https://api.muapi.ai/api/v1/predictions/{request_id}/result
# CLI
muapi runs get {request_id} --wait --output-jsonBeyond generation
Post or schedule to YouTube, TikTok, Instagram, Facebook, LinkedIn, Pinterest, Threads, and X. Connect an account once (OAuth, one-time), then publish through the same model workflow above.
Connect an account →Content moderation, OCR, speech transcription, and YouTube/TikTok/Instagram download & read tools — all discoverable and runnable the same way as any generative model.
See full list in SKILL.md →No subscriptions
One API key, one credit balance, pay per generation. Every model's cost is visible up front via estimate-cost — no plan tiers, no per-provider billing to manage.
Three ways to integrate
One line: point any skill-aware agent (Claude, custom LLM apps) at the file and it takes it from there.
set up https://muapi.ai/SKILL.mdView SKILL.md →Run muapi as an MCP server — every model becomes a tool call for Claude Code, Cursor, and other MCP clients.
muapi mcp serveView manifest →Terminal-first: discover, inspect, estimate, run, and poll from the shell — scriptable, --output-json everywhere.
pip install muapi-cliCLI docs →FAQ
What is a SKILL.md file?+
A machine-readable markdown file that teaches an AI agent how to use an API — which endpoints to call, in what order, and how to authenticate. Muapi's is published at muapi.ai/SKILL.md.
How does an agent pick which model to use?+
It calls GET /api/v1/models (optionally filtered by category or family) to get the live catalog, then GET /api/v1/models/{name} for full schema and pricing. No model list is ever hardcoded on either side.
Can an agent see the cost before running a model?+
Yes — POST /api/v1/models/{name}/estimate-cost with the intended payload returns the exact cost before any credits are spent.
Is there a subscription?+
No. One API key, one credit balance, pay per generation. Costs vary per model and are always visible via estimate-cost before you run anything.
Are provider names exposed to the agent?+
No — every response (catalog, schema, run, result) speaks only in Muapi model names, categories, and costs. The underlying provider is never surfaced.