Inference.Sh
Wire one MCP endpoint so your coding agent can browse and run 150+ hosted AI apps (image, video, audio, LLMs, 3D) without juggling separate API keys per vendor.
Overview
ac.inference.sh/mcp is a MCP server for the Build phase that lets coding agents browse and execute 150+ hosted AI apps and stream multimodal results over streamable HTTP.
What is this MCP server?
- Catalog of 150+ AI apps spanning image, video, audio, LLMs, and 3D generation
- Two streamable-http remotes: sh.inference.ac and api.inference.sh/mcp
- Browse apps from the agent, execute jobs, and stream results back into the session
- Server schema v2025-12-11; published as inference.sh MCP v1.0.1
- Reduces one-off SDK sprawl when prototyping multimodal features in a single repo
- 150+ AI apps in catalog
- 2 streamable-http remote URLs
- MCP server version 1.0.1
What problem does it solve?
Solo builders waste hours registering separate APIs for every image, video, audio, or 3D model they want the agent to call during feature work.
Who is it for?
Multimodal prototypes, content pipelines, and agent features that need on-demand generative APIs during active development.
Skip if: Teams that require only a single fixed LLM with no media generation, or builders who cannot use remote hosted inference for compliance reasons.
What do I get? / Deliverables
After you register the inference.sh remotes, your agent can discover apps, run jobs, and stream outputs in one MCP session instead of custom glue per modality.
- MCP tools to browse and execute inference.sh AI applications
- Streamed multimodal outputs inside the agent conversation
- Fewer per-vendor SDK integrations during build
Recommended MCP Servers
Journey fit
Solo builders add inference.sh during the build phase when they need generative or multimodal capabilities inside the agent loop rather than only in a separate web UI. integrations is the canonical shelf because the value is connecting Claude Code, Cursor, or Codex to remote streamable-http MCP remotes that execute external AI workloads.
How it compares
Unified remote inference catalog over MCP, not a repository-local agent skill or a static skills.sh install.
Common Questions / FAQ
Who is ac.Inference.Sh/mcp for?
Solo and indie developers using Claude Code, Cursor, Codex, or Windsurf who want one MCP connection to run many generative AI apps while building product features.
When should I use ac.Inference.Sh/mcp?
Use it in the build phase when you need the agent to execute image, video, audio, LLM, or 3D workloads and stream results without integrating each provider separately.
How do I add ac.Inference.Sh/mcp to my agent?
Add a streamable-http MCP remote pointing at https://sh.inference.ac or https://api.Inference.Sh/mcp in your client’s MCP server configuration, then restart the agent so tools load.