Skip to content
Moka the puppy, holding a coffee

Chat with any LLM. Plug in any MCP server. See every call.

Moka is a zero-config, 100% local sandbox for building, debugging and demoing AI agents, with a live inspector and generative UI built in. It can also show what your AI editor sends to your MCP servers.
Terminal window
npx @mokalabs/sandbox
Moka chat with tool calls and the live inspector

That’s it. Moka finds your API keys, detects Ollama, starts a built-in MCP server with demo tools, and opens a polished chat UI in your browser.

Any model

OpenAI, Anthropic, Gemini, Azure, Ollama, OpenRouter, Groq, DeepSeek, Mistral, xAI — or any OpenAI-compatible gateway with your own base URL and headers.

Any MCP server

stdio, Streamable HTTP and SSE. Paste your Claude Desktop, Cursor or VS Code config, or pick from the gallery.

Agent Skills

Drop in SKILL.md folders — including ~/.claude/skills — loaded on demand with progressive disclosure.

Inspector

Every model step, tool call, token and raw MCP JSON-RPC message, with timings and a per-run waterfall.

Generative UI

Render MCP Apps in a secure sandbox, and let any model answer with native A2UI forms, cards and buttons.

Proxy for your AI editor

npx @mokalabs/proxy wrap, and every message between GitHub Copilot, Cursor, Claude or Gemini and your MCP servers shows up live: timeouts, retries, errors.

Files, both ways

Attach PDFs and Word files, and local models read them too. Documents your tools make (PDF, Word, HTML, CSV…) open next to the chat.

Tool tests

Prompts with the tools a model should call, run on several models at once, with a scorecard. moka test runs them in CI.

Security check

Tool poisoning, invisible characters, shadowing and tools that change after you trusted them, flagged as servers connect.

Compare models

One prompt, several models, the same tools. Answers, latency, tokens and tool calls side by side.

Your agents too

Point Moka at a LangGraph, CopilotKit or ADK agent over AG-UI or A2A, and get the chat, A2UI and the inspector for it.

100% local

Runs on your machine. No account, no telemetry, no Moka backend. Prompts go only to the model you pick, and with Ollama nothing leaves your machine.

Ship it

Export a workspace to Vercel AI SDK or LangGraph code, share a moka.json, or run the Docker image for your team.