korkyzer/multica-mcp

开发工具

本页由 AI 翻译，技术细节请以原文为准。查看原文

28 stars 7 forks 质量 76 趋势 15MIT

Repository Install

一个MCP服务器（stdio，TypeScript）包裹了Multica的CLI，这样LLM聊天就能直接驱动Multica：打开问题、分配代理、跟踪进度、串联工作流程。

概览

一个MCP服务器（stdio，TypeScript）包裹了Multica的CLI，这样LLM聊天就能直接驱动Multica：打开问题、分配代理、跟踪进度、串联工作流程。 Multica CLI是主要工具。它涵盖了MCP涵盖的所有内容，甚至更多，非常适合大多数工作流程。这里没有任何内容旨在取代它。我发现Claude Desktop（常规聊天）是头脑风暴和架构讨论的最佳环境，但我需要它能打开Multica问题、分配代理、跟踪进度，并直接从对话中串联连续工作流程，而无需切换到终端。这就是我构建这个MCP的原因。如果你把思考时间花在大型语言模型聊天（Claude Desktop、Codex Desktop 或任何支持MCP的客户端）里，这让聊天本身能与Multica通信。如果你住在终端里，继续使用CLI。

概述（AI 翻译）

一个MCP服务器（stdio，TypeScript）包裹multica的CLI，使LLM聊天能直接驱动Multica：开问题、分配代理、跟踪进度、链式工作流程。

为什么会有这样的存在

Multica CLI是主要工具。它涵盖了MCP涵盖的所有内容，甚至更多，非常适合大多数工作流程。这里没有任何内容旨在取代它。

这个MCP针对一种特定情况：

我发现Claude Desktop（常规聊天）是脑力激荡和架构讨论的最佳环境，但我需要它能直接打开Multica问题、分配代理、跟踪进度，并直接从对话中串联顺序工作流，而无需切换到终端。这就是我构建这个MCP的原因。

如果你把思考时间花在大型语言模型聊天（Claude Desktop、Codex Desktop 或任何支持MCP的客户端）里，这让聊天本身能与Multica通信。如果你住在终端里，继续使用CLI。

具体用例

在聊天中头脑风暴，然后在不离开对话的情况下将具体方案交给Multica的代理人
从聊天中运行冲刺规划：起草议题、设定优先级、分配代理、排队跟进

本页由 AI 翻译，技术细节请以原文为准。

README

以下为原文 README 节选，完整内容请查看原文。

multica-mcp

An MCP server (stdio, TypeScript) that wraps the multica CLI so an LLM chat can drive Multica directly: open issues, assign agents, follow progress, chain workflows.

Why this exists

The Multica CLI is the main tool. It covers everything this MCP covers, plus more, and it is the right fit for most workflows. Nothing here is meant to replace it.

This MCP is for one specific situation:

I find Claude Desktop (regular chat) is the best environment for brainstorming and architecture discussions, but I needed it to be able to open Multica issues, assign agents, follow progress, and chain sequential workflows directly from the conversation, without switching to a terminal. That is why I built this MCP.

If you spend your thinking time in an LLM chat (Claude Desktop, Codex Desktop, or any MCP-capable client), this lets the chat itself talk to Multica. If you live in the terminal, keep using the CLI.

Concrete use cases

Brainstorm in chat, then hand off concrete pieces to Multica agents without leaving the conversation
Run a sprint planning from the chat: draft issues, set priorities, assign agents, queue follow-ups
Review agent results from chat: pull run messages, read comments, comment back, re-open or re-assign

Requirements

multica installed on PATH and authenticated
Multica daemon running
Node.js 20+
pnpm

Build

cd /path/to/multica-mcp && pnpm install && pnpm build

The bundle is emitted at dist/server.js.

Run

cd /path/to/multica-mcp && pnpm start

Logs

Tool calls are logged as JSONL with timestamp, tool, params, duration_ms, success, and error when relevant. The log file is resolved in this order:

MULTICA_MCP_LOG_PATH environment variable (absolute path)
$XDG_STATE_HOME/multica-mcp/calls.log
~/.local/state/multica-mcp/calls.log

Configuration

Env var	Purpose
`MULTICA_APP_URL`	Base URL used when building issue and project URLs
`MULTICA_MCP_LOG_PATH`	Absolute path override for the JSONL log
`XDG_STATE_HOME`	Standard XDG base dir for the default log location

When MULTICA_APP_URL is unset, the server reads multica config show and, if nothing matches, returns a null URL (no localhost fallback).

Tools

`multica_list_agents`

Returns [{ id, name, provider, model_hint, description }].

{}

`multica_list_projects`

Returns [{ id, name, description, issue_count }].

{}

`multica_create_project`

Returns { id, name, url }.

{
  "name": "Onboarding revamp",
  "description": "Agent coordination project for the onboarding revamp."
}

`multica_create_issue`

Returns { id, short_id, title, status, assignee, url }.

{
  "title": "Add a login flow",

在 GitHub 查看完整 README

安装

This server does not publish a one-line install command.

Open the repository installation guide 新手？先看 mcpskill.org 教程 →

配置

{
  "mcpServers": {
    "multica": {
      "command": "node",
      "args": ["/absolute/path/to/multica-mcp/dist/server.js"]
    }
  }
}