Gradio 5.38 adds local file support and live progress tracking for MCP servers

The latest Gradio release solves one of the most annoying friction points for anyone using MCP servers with LLM agents: the file upload headache. Before version 5.38.0, if you pointed a remote Gradio MCP server at a file on your machine, you’d get an error. The server couldn’t reach your local image or audio file because it expected a public URL. The workaround meant manually hosting files somewhere first — a step that feels absurd when you’re trying to build autonomous agent workflows.

The team’s fix is pragmatic. Gradio now ships a dedicated “File Upload” MCP server that agents can use to push files directly to the application. The connection docs automatically surface instructions for starting it, and the guides include security considerations worth reading before you expose anything to the open web.

Alongside local files, Gradio 5.38 introduces real-time progress streaming. Long-running AI tasks no longer leave your MCP client staring at a void. The server emits status updates as the work progresses. Developers who build custom MCP tools should implement these progress hooks — the guide explains how, and doing so will make your tools feel far more responsive inside clients like Claude Desktop.

The release also targets backend integration pain. A new gr.load_openapi function converts an OpenAPI spec into a Gradio app with MCP tooling in a single line of code. That replaces the manual, error-prone chore of mapping endpoints to MCP tools. Authentication gets smarter too: typing arguments as gr.Header lets Gradio automatically extract headers from incoming requests and display them in the MCP connection docs, so users know exactly which credentials to supply. And for tool descriptions, the api_description parameter lets you override auto-generated text — useful when your function name doesn’t tell the full story. These are incremental changes, but each one removes a small, persistent annoyance from the MCP development loop.