Llama.cpp WebUI Review: Context Tracking and MCP Proxy Tips
Use llama.cpp's webui for better context tracking and implement MCP filtering to avoid context blowup.
Implement MCP proxy with filesystem filtering as shown in the gist to reduce context usage.
Summary
A user has compared several chat UI frameworks and prefers the llama.cpp webui for its precise context counter that displays token usage before and after each response. The interface eliminates the guesswork of hidden context limits, which can otherwise cause the model to become incoherent. However, the UI currently fails when a tool call fails, terminating the entire conversation, and lacks support for folders, workspaces, or project‑level system prompts. The user also notes the absence of MCP tool controls, which would allow selective tool exposure to the client. A custom MCP proxy script is provided that filters filesystem tool calls and reduces context usage by excluding unwanted directories. The script can be launched with a single command and integrates seamlessly with the webui. This review highlights both the strengths and gaps of the llama.cpp UI for developers building local LLM applications.
Key changes
- Precise token counter before and after each response
- Conversation ends when a tool call fails
- No support for folders, workspaces, or project prompts
- Missing MCP tool controls for selective tool exposure
- Custom MCP proxy filters filesystem calls to reduce context
- Proxy script launched with a single command
- Integrates seamlessly with the webui
- Improves context usage and stability