Pic2SVG MCP

Pic2SVG MCP is the agent-facing interface to pic2svg, an image-to-SVG tracing service operated by Song Zhi Jun. A remote MCP server at the pic2svg.com/api/mcp endpoint exposes one tool, trace_image_to_svg, and a public trace API offers the same engine over HTTP. Neither requires an account or any local install today.

What It Is

Pic2SVG MCP lets an AI agent send an image as base64 and receive SVG text back, along with the trace mode, canvas size, path count and file size. The tool accepts an optional file name and a preset of sharp, balanced or compact. It works with any MCP client that supports remote servers over HTTP. The site says a local stdio server and command-line tool (the pic2svg-mcp package) are coming soon and are not yet on npm.

How the workflow runs

A user adds the remote server URL to their MCP client configuration, then asks the agent to trace a file, for example a logo PNG with the compact preset. The agent reads the file, calls the tool and gets SVG text, which it can save under a name such as name.pic2svg.svg so an existing SVG is not overwritten. Scripts can instead make a single HTTP POST to the trace API, which returns JSON containing svgText, mode, canvas and sizeKB.

Engine behavior and data handling

The engine is deterministic rather than generative: the same input and settings give the same paths. It is built on a vtracer-based tracer. Transparent areas become white shapes, text becomes outlined shapes, and larger images are scaled down to about 600,000 pixels before tracing. Only SVG is produced through the API and MCP server. Images are uploaded to pic2svg's own server and, per the site, are not sent to third-party AI services; temporary trace files are deleted when the trace finishes. The API has a daily per-client allowance and a request size limit of about 45 MB.