What it does
A terminal client that bridges local Ollama models (or Ollama Cloud) to one or more MCP servers, providing interactive access to MCP tools, prompts, and resources with real-time streaming. Supports agent mode for multi-step tool execution with configurable loop limits, human-in-the-loop approval gates, MCP server hot-reload for rapid development, and vision-capable model support.
Who it's for
Backend engineers and MCP server developers testing tools locally, harness engineers debugging agent workflows with safety controls, and developers prototyping tool-calling agents before moving to production systems.
Common use cases
- Test and validate MCP tools, prompts, and resources locally before deploying to production
- Debug multi-step agent workflows with human-in-the-loop approval on each tool execution
- Rapidly iterate on MCP servers using hot-reload without restarting the client
- Prototype agents with local Ollama, then switch to Ollama Cloud models for testing
- Experiment with model parameters, system prompts, context windows, and reasoning modes interactively
Setup pitfalls
- Requires Ollama running locally or valid Ollama Cloud credentials; does not fetch or pull models automatically
- Writes configuration and conversation history to disk — verify write permissions and available storage
- Classified as high-risk due to filesystem access and network calls to MCP servers — isolate untrusted servers or use in a sandbox
- Tool execution blocks the terminal UI; resource-intensive tools may freeze the interface until they complete