One MCP Server.
160+ AI Models.
Connect Claude Desktop, Cursor, VS Code, or Windsurf to RunAPI with Hosted MCP and sign in through OAuth.
https://mcp.runapi.ai/mcphttps://mcp.runapi.ai/mcphttps://mcp.runapi.ai/mcphttps://mcp.runapi.ai/mcpHosted MCP uses Streamable HTTP at https://mcp.runapi.ai/mcp. Add this URL in Claude Desktop, Cursor, VS Code, or Windsurf, then complete the RunAPI OAuth flow in your browser. Use the separate API Key fallback only for compatibility or headless environments.
API Key fallback
Use these configurations only when a client cannot complete OAuth or in a headless environment. Keep the API key in the client's secret mechanism instead of writing it directly into JSON.
{
"mcpServers": {
"runapi": {
"url": "https://mcp.runapi.ai/mcp",
"headers": {
"Authorization": "Bearer ${env:RUNAPI_API_KEY}"
}
}
}
}macOS GUI: Run the command below in Terminal, then fully quit Cursor with Cmd+Q and reopen it from Finder or Dock. Repeat after logging out or restarting macOS. This exposes the key to apps started later in your login session, so use a dedicated revocable key and run `launchctl unsetenv RUNAPI_API_KEY` when you no longer need it.
printf 'RunAPI API key: ' >&2
read -r -s RUNAPI_API_KEY
printf '\n' >&2
launchctl setenv RUNAPI_API_KEY "$RUNAPI_API_KEY"
unset RUNAPI_API_KEY{
"mcpServers": {
"runapi": {
"serverUrl": "https://mcp.runapi.ai/mcp",
"headers": {
"Authorization": "Bearer ${file:~/.config/runapi/mcp_api_key}"
}
}
}
}macOS/Linux: Run the command below once to create a personal mode 600 key file. Save the copied JSON, then fully quit and reopen Windsurf so it rereads the configuration. This avoids an empty GUI environment variable.
install -d -m 700 ~/.config/runapi
umask 077
printf 'RunAPI API key: ' >&2
read -r -s RUNAPI_API_KEY
printf '\n' >&2
printf '%s' "$RUNAPI_API_KEY" > ~/.config/runapi/mcp_api_key
unset RUNAPI_API_KEY
chmod 600 ~/.config/runapi/mcp_api_key{
"inputs": [
{
"type": "promptString",
"id": "runapi-api-key",
"description": "RunAPI API key",
"password": true
}
],
"servers": {
"runapi": {
"type": "http",
"url": "https://mcp.runapi.ai/mcp",
"headers": {
"Authorization": "Bearer ${input:runapi-api-key}"
}
}
}
}What is RunAPI MCP Server?
RunAPI MCP Server connects MCP-compatible coding tools to RunAPI. Your AI assistant can browse models, inspect input parameters, check pricing, search reusable prompts, create media tasks, poll task status, and check account balance from the editor. Claude Desktop, Cursor, VS Code, and Windsurf use OAuth; API Key authentication remains available separately for compatibility and headless environments.
8 Built-in Tools
Everything your AI assistant needs to discover, price, and run AI models.
list_models
Browse models by modality, service, or endpoint.
get_model_info
Inspect input params, constraints, and pricing for a model.
list_actions
Group endpoint names by output modality.
check_pricing
Return pricing snapshot for a service, endpoint, and model.
search_prompts
Search reusable prompt examples by modality, category, tags, or text.
create_task
Create a media task and optionally poll until completion.
get_task
Fetch status and latest payload for an existing task.
check_balance
Return account balance and spending metrics.
Optional local installation
Local MCP is a separate stdio option. Use it with Node.js and Local Login when you prefer a local process; Hosted MCP above does not require one.
Start the local stdio server with Node.js.
npx -y @runapi.ai/mcp
Run `runapi login` to open browser-based Local Login and save credentials to local config.
runapi login
In headless environments, set RUNAPI_API_KEY before starting the local process.
RUNAPI_API_KEY=runapi_xxx npx -y @runapi.ai/mcp
160+ Models, 4 Modalities
Every listed model is executable through this MCP server.
| Modality | Models |
|---|---|
| Image | Flux Kontext, Nano Banana, GPT Image, Seedream, Ideogram, Qwen |
| Video | Kling, Veo, Seedance, Hailuo, Wan, Runway, Luma |
| Audio | Suno Music, ElevenLabs TTS, Sound Effects, STT, Isolation |
| Utility | Kling AI Avatar, OmniHuman Detection, Seedream Layer Decomposition |
Frequently Asked Questions
Claude Desktop, Cursor, VS Code, and Windsurf complete RunAPI OAuth in the browser. API Key authentication remains a separate fallback for compatibility and headless environments.
Yes. Local MCP runs as a separate stdio process with Node.js. Use Local Login interactively or set RUNAPI_API_KEY for headless hosts.
Explore More
Ready to connect your editor to 160+ AI models?
Install the MCP server and start generating in minutes.