Tools / Model Context Protocol
Useful tools.
Understandable boundaries.
MCP gives an AI client a structured way to use tools. These projects explore what happens when capability, permission, and a visible result are designed together.
01 / Hosted & open source
TurtlePen
Give an agent an exact drawing language, text measurement, semantic timelines, and a way to check the output. The full hosted engine exposes 87 tools; a smaller browser companion lets you work together on the page.
The hosted engine also supports physical 3D geometry, curved sweeps, lofts, and TPF, STL, and GLB export. Explore 3D geometry.
02 / Local & open source
Local Files MCP
A Python MCP server with a GUI installer for constrained local-file access. Setup happens on your computer; the tool’s permissions and actual connection are part of the workflow.
Explore the implementation and setup documentation in the original repository. It runs locally and is not a file-hosting service on this website.
The surrounding work
Tools need an operating context.
SCANGATE
Examine outside artifacts before a working agent can read them.
Agent Enablement Framework
Give people and assistants a shared set of project contracts and workflows.
Neuron
Bring source work, agent execution, decisions, and proof into a local workbench.
RegistryRouter research
An earlier tool-routing research package in the public site repository. A research artifact, not a hosted endpoint.