Runware

Every Runware model,
inside your agent.

Connect your agent to the Runware MCP server.

Claude, Cursor, Codex, and any MCP client reach the whole Runware catalog. Across all modalities. Name your models, or let the agent pick them for you.

Get an API keyRead the docs
ONE URL · ANY MCP CLIENT · OAUTH 2.1 · NO SURCHARGE
Connecting Runware
  1. Adding mcp.runware.ai connector
  2. Signing in with OAuth
  3. Tools ready in your agent
How it works

One connection. The whole catalog.

Point any MCP-compatible agent at Runware's hosted server and it gets a set of tools. From there it generates, searches the catalog, and checks pricing on your behalf. Every Runware-hosted model is one tool call away.

1

Connect

Add Runware as a custom connector in any MCP client and sign in once with OAuth. The hosted server needs nothing installed on your machine.

2

Ask

Describe what you want across image, video, audio, 3D, or text. The agent picks a model, or uses the one you name, and fills in the parameters. No task types or schemas to look up.

3

Get

The agent runs the model, hands back the result, and reports exactly what it cost.

Agents

Works in the agents you already use.

Connect any MCP-compatible client, two ways: the hosted server, or a package you run yourself.

ClaudeClaude CodeCursorVS CodeCodex+ any MCP client

Hosted

https://mcp.runware.ai

Point any client at the hosted server. Nothing to install.

no endpoint

Local

npx -y @runware/mcp

Run the server yourself. For air-gapped or locked-down networks.

Why the agent

The API you have, with the decisions made for you.

The MCP adds the decisions you'd write yourself. Either tell it your preferred model or let it pick one for you. It then fills the parameters and validates them against the schema before it runs.

RESTPOST /v1

You write this.

POST https://api.runware.ai/v1
{
  "taskType": "imageInference",
  "taskUUID": "5a6f60a0-02a9-4630-91ae-9d83426af3b1",
  "model": "runware:400@1",
  "positivePrompt": "hero shot of the new running shoe, studio light",
  "width": 1024,
  "height": 1536,
  "steps": 32,
  "CFGScale": 4
}
Look up the model, the schema, the params.
MCP

The agent handles it.

You

Hero shot of the new running shoe, portrait, studio light.

Runware MCP

  • picks runware:400@1
  • fills the parameters
  • validates the schema
  • runs it
You describe the outcome. That's it.
Multi-modal chain

Chain models in one request.

The agent doesn't stop at one generation. Give it an outcome and it chains models across modalities on its own.

Take this image and make it a 8-second video. Add native audio using Mirelo SFX 1.5. Then add a voiceover at the end that says "Start your drum lessons today"
Models

The whole library, one call away.

Image, video, audio, 3D, and text models from every major provider, plus thousands of community models. The agent reaches any of them through one tool, addressed by a single identifier.

400K models across every modality

See full model library

USE CASES

Where it earns its place.

The MCP is most useful where you already work, without a tab switch to a separate generation tool.

Inside your coding agent

Build the UI and generate the whole asset set in the same session, stills and a short clip, no export, no second tool.

You

Generate a hero image, two product shots, and an 8-second promo clip for this landing page. 1:1 stills, studio light.

RUNWARE MCP

  • run · google:4@3 · 1080x1080 (hero)$0.06895
  • run · google:4@3 · 2 product shots$0.14
  • run · bytedance:seedance-2-0 · 8s clip, 1080p$1.26801

Four assets in your repo, $1.48 on your balance.

Hero image for landing page
Product shot one
Product shot two
8s
Built for production

Built to run on your account.

Your key stays server-side.

OAuth wraps your key on the server and keeps it encrypted. The client only ever holds a token.

No surcharge.

MCP calls cost what the underlying task costs, billed to your own balance. Every other tool is free.

Same tools as production.

Search the catalog, check pricing, and preview sample outputs — the same MCP tools you'd wire into a pipeline.

FAQ

Questions, answered.

An open standard that lets AI agents call external tools. Runware exposes its API as MCP tools, so any MCP-compatible agent can drive Runware directly.

Runware MCP

Connect Runware to your agent.

Get a key, paste one URL, and start generating.

Get an API keyRead the MCP docs