Moka Chat gives Ollama models MCP tools via one npx command, with an inspector showing every tool call
A developer released Moka, a free, MIT-licensed local sandbox for testing prompts and MCP servers visually. It runs via 'npx @mokalabs/sandbox' with zero config, auto-detects running Ollama instances, and includes an inspector exposing raw JSON-RPC traffic, tokens, and latency. Developers can test tool calls and streaming, compare models side by side, and export working LangGraph Python scripts directly.
- Moka is an open-source, MIT-licensed, local-only chat app for testing models with MCP servers, auto-detecting a running Ollama instance on start
- One command, npx @mokalabs/sandbox, launches it; MCP servers (filesystem, fetch, git, Playwright, custom) are added from a gallery or by pasting Claude Desktop / Cursor configs
- The inspector displays every tool call with raw JSON-RPC, token counts, and timings, showing why a model did or didn't call a tool
- Moka also works with LM Studio, Anthropic, OpenAI, and other hosted models for side-by-side comparison on the same tools
- Moka is 100% free with no paywalls, MIT-licensed on GitHub at github.com/mokahq/mokalabs
- An Export-as-code feature generates a working LangGraph Python script from tested tool calls and prompts, and Moka can debug existing LangGraph agents over AG-UI
Related
Developer switches from Ollama to LM Studio for local LLMs on Windows
A developer details their migration from Ollama to LM Studio for running local LLMs on Windows, citing faster inference via llama.cpp, VRAM-fit badges, GUI parameter control, MCP integrations, and the ability to reuse existing Ollama models. The post reflects growing community interest in LM Studio's features for local AI workflows.
A developer details their migration from Ollama to LM Studio for running local LLMs on Windows, citing faster inference via llama.cpp, VRAM-fit badges, GUI parameter control, MCP integrations, and the ability to reuse existing Ollama models. The post reflects growing community interest in LM Studio's features for local AI workflows.
Local pass-through proxy for inspecting MCP traffic between Copilot agent mode and MCP servers
A local CLI proxy that sits between GitHub Copilot agent mode and MCP servers, logging every message, payload, latency, timeout, and retry so developers can debug tool calls that are otherwise invisible. It wraps an existing VS Code mcp. config and can be unwrapped to restore it. Runs entirely locally with zero config.
A local CLI proxy that sits between GitHub Copilot agent mode and MCP servers, logging every message, payload, latency, timeout, and retry so developers can debug tool calls that are otherwise invisible. It wraps an existing VS Code mcp. config and can be unwrapped to restore it. Runs entirely locally with zero config.
Eidon: self-hosted all-in-one AI platform with chat, agents, automations in one Docker container
Eidon is a self-hosted AI chat and agent platform packaged as a single Docker container, with web research, MCP, skills, browser use, image generation, and automations built in. LLMs power its chat, Grok-bot-like agents, and automations, and it works with local models via Ollama or LM Studio. It targets self-hosters who want these capabilities without stitching together an app plus sidecars and external tools.
Eidon is a self-hosted AI chat and agent platform packaged as a single Docker container, with web research, MCP, skills, browser use, image generation, and automations built in. LLMs power its chat, Grok-bot-like agents, and automations, and it works with local models via Ollama or LM Studio. It targets self-hosters who want these capabilities without stitching together an app plus sidecars and external tools.
MCPFlo: deterministic testing and debugging tool for MCP servers
MCPFlo is an offline-first, open-source Electron app for testing and debugging MCP servers. It replaces the common practice of relying on an LLM to judge server behavior with deterministic Chai.js-style assertions against real tool responses. It auto-generates nested schema forms via RJSF for easy input.
MCPFlo is an offline-first, open-source Electron app for testing and debugging MCP servers. It replaces the common practice of relying on an LLM to judge server behavior with deterministic Chai.js-style assertions against real tool responses. It auto-generates nested schema forms via RJSF for easy input.
How much of an AI agent is the framework, and how much is just a loop?
A developer built a shopping assistant agent as a raw tool-calling loop in plain Python, then rebuilt it with LangGraph's create_agent to compare. The experiment shows the framework mainly runs the same loop plus adds memory, streaming, retries, and tracing. The author recommends writing the loop once before adopting a framework to keep agent behavior debuggable.
A developer built a shopping assistant agent as a raw tool-calling loop in plain Python, then rebuilt it with LangGraph's create_agent to compare. The experiment shows the framework mainly runs the same loop plus adds memory, streaming, retries, and tracing. The author recommends writing the loop once before adopting a framework to keep agent behavior debuggable.