Ido Salomon

Interactive tools over MCP

MCP-UI expanded the interaction between an AI assistant and a tool beyond a plain text response. Its server packages create UI resources, while client components render them in a sandbox and support communication between the embedded interface and the host. The project describes these patterns as an influence on the MCP Apps specification and now implements that standard. Salomon developed the project in collaboration with Liad Yosef. The contribution is an interface layer: a tool can return a usable control surface alongside information for the model.[2]

Documentation as a tool

GitMCP exposes a GitHub repository or GitHub Pages site through a remote MCP server. An assistant can search the project’s documentation and code instead of relying only on what its model learned during training. A repository-specific endpoint keeps requests focused on one project; a generic endpoint supports switching between repositories but depends on selecting the right one. The repository includes connection instructions for multiple editors and assistants, as well as self-hosting options.[3]

AgentCraft borrows controls from real-time strategy games to display agents, launch work and manage their lifecycle. It combines a shared view of ongoing activity with controls for approvals and messages. The product also offers isolated containers, remote access and a shared room for coordinating agents across machines. Salomon’s underlying design question is how people can supervise more work without turning every agent into another conversation to monitor. His public profile lists Tel Aviv University studies from 2014 to 2018 and describes both electrical-engineering and computer-science capstone projects.[4][5]