Skip to main content
A Skill is a set of instructions, with any supporting files and scripts, that your AI client reads from Biohub MCP and follows step by step. Biohub MCP serves a Skill’s files as text and never runs them. Any command a Skill asks for runs wherever your client runs it, usually your own computer. Like the rest of Biohub MCP, Skills need no sign-in.

​
Choose a Skill

​
WaveOrder

WaveOrder is an open-source package from the Mehta Lab at Biohub that makes physics-based reconstructions of label-free and fluorescence microscopy data. Its source code and examples are on GitHub.

Reconstruct microscopy data

Go from a raw stack or a sample dataset to a tuned reconstruction config.

Visualize a transfer function

Compute a config’s transfer function and inspect it in napari.

​
What your client needs

The WaveOrder Skills call no Biohub tools. They need a client that can run commands on your computer, such as Claude Code, Codex, or Cursor, with uv installed and a display for napari. A chat-only client can read these Skills but can’t follow them. Review each command before you approve it, as Security and privacy explains.

​
Start a Skill

Ask for the task and name the Skill, such as “Using the Biohub MCP waveorder-tutorial Skill, …”. When a request matches a Skill, the server’s instructions tell your client to read that Skill’s SKILL.md first. A client that can’t read MCP resources can call read_resource with the SKILL.md URI, such as skill://waveorder-tutorial/SKILL.md, and list_resources to see every Skill file. Some clients, such as Codex, ask you to approve read_resource first. If your client answers without the Skill, see The assistant ignores a Skill.

​
Files and versions

Every Skill serves these resources under skill://<name>/: Each Skill resource also carries the Skill’s version and every file’s size and SHA-256 hash in its MCP _meta, under biohub.ai/skill-provenance. Use that live metadata, not a value copied from these docs, when you need the exact deployed version.

​
Next steps