Skip to content
Moka

Introduction

Moka is a local-first sandbox for AI agents. Run one command and you get a polished web app where you can chat with any LLM, connect MCP servers and Agent Skills, watch every step in a live inspector, and render interactive UI from tools and models.

Who it’s for

Section titled “Who it’s for”

Engineers building agents

Try a new MCP server, model or prompt in seconds. See the exact JSON-RPC traffic, tool arguments, token counts and latencies. Export working code when you’re done.

MCP server authors

Test your tools, resources, prompts and MCP Apps against real models — and call tools directly without a model in the loop.

Anyone using an AI editor

Put the Moka proxy in front of the MCP servers of GitHub Copilot, Cursor, Claude or Gemini, and see exactly what they send: timeouts, retries and errors included.

Solution & sales engineers

Build repeatable demos with workspaces, starter prompts and presenter mode. Share a moka.json and anyone can reproduce your demo with npx.

Platform teams

Point Moka at your internal LLM gateway and MCP servers, ship the Docker image, and give every team a sanctioned playground.

How it works

Section titled “How it works”
Browser UI  ──HTTP/NDJSON──▶  Moka server (Hono)  ──▶  @mokalabs/core
 chat · compare · tools                                  ├─ providers   (AI SDK: OpenAI, Anthropic, Gemini, Azure, OpenAI-compatible)
 settings · inspector                                    ├─ MCP manager (stdio · Streamable HTTP · SSE, raw JSON-RPC tap)
 MCP Apps · A2UI renderer                                ├─ skills      (SKILL.md, load_skill / read_skill_file)
                                                         ├─ agent loop  (streaming tool calls, render_ui)
                                                         └─ event bus   (everything the inspector shows)
  • 100% local. Moka runs on your machine (or your container). No account, no telemetry, no Moka backend. Your keys and data go only to the providers and servers you configure, and with a local model nothing leaves your machine. Details
  • One config file. Everything you set in the UI is saved to a readable moka.json. Secrets stay in environment variables via env:NAME references.
  • Standards, not lock-in. MCP for tools, Agent Skills for instructions, MCP Apps and A2UI for UI, the Vercel AI SDK for providers. What you build in Moka works elsewhere.

Packages

Section titled “Packages”
PackageWhat it is
@mokalabs/sandboxThe app you run: CLI, server, web UI and a demo MCP server
@mokalabs/coreThe headless engine, if you want to embed Moka in your own tools
create-mokanpm create moka — scaffolds a shareable demo project