This skill uploads local assets to a Stitch project. A Python script reads the file and posts it to the Stitch API, which avoids the output token limit that stops a model from sending base64 data through MCP tools.
It supports PNG, JPEG, WebP, HTML and Markdown files, sets screen titles and records which tool generated the file. The agent asks for approval before it runs the upload.
What You Get
- Screens created in a Stitch project from local images and HTML files
- A DESIGN.md uploaded as a design markdown screen
- Screen titles set from a title option such as a route path
- A generated-by tag that records the source of each upload
Common Use Cases
- Uploading extracted HTML from a local app into Stitch
- Adding a mockup image to a project before editing it
- Sending a DESIGN.md that is too large for a direct MCP call
What Makes It Unique
It solves a concrete limit: a model cannot output the base64 of even a small file through MCP. The script encodes the file in-process, detects the MIME type from the extension, and includes a fix for macOS certificate errors.
Under The Hood
The agent reads the MCP configuration file of the host tool, such as ~/.claude.json or the Gemini settings, to find the Stitch API key, and asks the user when it cannot. It then runs a Python script that base64-encodes the file and sends it to the Stitch batchCreate endpoint. A user checkpoint comes before the run.