Replicate MCP 🔌 MCP server Freemium

What it is

Replicate MCP is Replicate's official Model Context Protocol server. It exposes Replicate's HTTP API, so an assistant can search models, run predictions and read model metadata. It is offered as a hosted server at mcp.replicate.com with web sign-in, and as the replicate-mcp npm package you run locally.

Key features

• Search Replicate's catalogue by task (for example, upscalers or image models)
• Run predictions and fetch outputs
• Read model metadata and input schemas
• Covers all operations of Replicate's HTTP API
• Experimental code mode: the agent writes TypeScript that runs in a Deno sandbox to chain several calls

How to install

Create a token at replicate.com/account/api-tokens. For Claude Code, Replicate's docs give claude mcp add "replicate-code-mode" --scope user --transport stdio -- npx -y replicate-mcp@alpha --tools=code (code mode needs Node.js and Deno). Claude Desktop, Cursor and VS Code Copilot use the same package with REPLICATE_API_TOKEN set, or connect to mcp.replicate.com. Note that the stable npm tag lags far behind the alpha track, so pin the version the docs recommend.

Pricing and rights

The server is free. Model runs are billed by Replicate per second of hardware or per output, depending on the model, with no monthly plan. Output rights depend on each model's licence, which Replicate shows on the model page.

Who it is for

Developers and technical creators who want open models (image, video, audio, upscaling) inside an agent, or who already host their own models on Replicate.

Verdict

Broad model coverage and an official source make it a solid media server. The version split between stable and alpha builds, and the experimental code mode, add setup friction for non-developers.