by Magiczerowxy
An integrated chat page inside Magpie's window that uses Magpie's configured models, instructions, skills and MCP servers, supporting multimodal input, streaming responses, LaTeX rendering, and persistent conversations.
Magpie Chat adds a single Chat page to the Magpie application. The page runs inside Magpie’s sandboxed UI and reuses the models, token limits, instructions, skills and MCP services that are already configured for Magpie, eliminating the need for a separate agent.
magpie plugin add Magiczerowxy/magpie-chat
or through Plugins → Discover → Unofficial · GitHub and search for magpie-chat./v1/models; models with reasoning levels expose a “thinking effort” selector.~/.config/magpie/library/ in the right‑hand settings pane.~/.config/magpie/magpie-chat/conversations/ and can be pinned or deleted.tool_calls execution for registered MCP servers.image_url, thumbnails shown in bubbles.$…$, $$…$$, \(...\), \[...\] via bundled KaTeX, no CDN.Q: Do I need to create a new account or provider for Magpie Chat? A: No. The plugin reuses the credentials and quota already configured in Magpie; it never stores additional credentials.
Q: Where are my chats saved?
A: Under ~/.config/magpie/magpie-chat/conversations/, each conversation is a separate JSON file.
Q: Can I use custom models not listed by the gateway?
A: Only models exposed through Magpie’s gateway (/v1/models) are selectable.
Q: Are there limits on image size or tool‑call rounds? A: Images are base64‑encoded with a 24 MB per‑image limit. Tool calls are capped at 8 rounds per request.
Q: Does the plugin require a build step? A: No. All files in the repository are loaded directly by Magpie.
在 magpie 窗口里加一个「对话」页面。 用 magpie 已经接好的模型、指令、技能和 MCP 服务器聊天,不用再开一个 Agent。
A chat page inside magpie's window. Talk with magpie's own models, instructions, skills and MCP servers — no second agent needed.
/v1/models,切换即时生效;有推理档位的模型会给出对应的思考强度选项。~/.config/magpie/library/ 的指令与技能(拼进 system prompt)、勾选 MCP 服务器(工具以 tool_calls 真正执行)。image_url 交给模型,气泡里显示缩略图、点开看大图。$…$、$$…$$、\(…\)、\[…\] 由本地 KaTeX 渲染,随包打包,不连 CDN。magpie-chat/conversations/;每条可以置顶或删除。在 magpie 窗口里的一次真实对话:公式由本地 KaTeX 渲染,表格正常排版,右侧设置栏可以调会话区与输入框宽度。

magpie plugin add Magiczerowxy/magpie-chat
或者在应用里:插件 → 发现 → 非官方插件 · GitHub,搜索 magpie-chat 后点「安装」。
装好后侧栏「插件」里点「对话」即可。
这个插件不注册供应商、不需要登录、不保存任何凭据。它复用 magpie 自己已经配好的模型和额度,请求全部由 magpie 的网关发出,所以用量统计与故障切换和 Agent 一致。
它只保存你自己的东西,都在 magpie 配置目录下:
~/.config/magpie/magpie-chat/
├── settings.json 模型、思考强度、系统提示词、勾选的指令/技能/MCP、窗口布局
└── conversations/ 每个对话一个 JSON
magpie 的插件页面跑在 sandbox="allow-scripts" 的框架里:没有网络权限、不能内嵌框架、不能用存储,唯一允许的联网动作是把自身导航到别处。所以这个插件是两层的:
page/index.html 是 magpie 托管的「带路页」,读窗口主题后 location.replace 到本地服务。index.mjs 作为 OpenCode v1 插件被 magpie 的 Bun 宿主加载,加载时启动那个本地服务(127.0.0.1:3477)。lib/server.mjs 提供 /api/* 并把聊天请求转发给 magpie 网关;chat/ 是真正的界面。服务只监听回环地址,并拒绝非同源请求(只接受 opaque 的 Origin: null 与自身 origin),任意网页都无法借它消耗额度。
/v1/models; switching takes effect immediately, and models with reasoning levels get a thinking-effort picker.~/.config/magpie/library/ (folded into the system prompt) and MCP servers (their tools run as real tool_calls).image_url; thumbnails appear in the bubble and open full size.$…$, $$…$$, \(…\) and \[…\] render through a bundled KaTeX — no CDN.magpie-chat/conversations/ in magpie's config directory; each can be pinned or deleted.A real conversation inside magpie's window: math rendered by the bundled KaTeX, tables laid out normally, and the settings sidebar on the right for adjusting conversation and composer width.

magpie plugin add Magiczerowxy/magpie-chat
Or in the app: Plugins → Discover → Unofficial · GitHub, search magpie-chat, click Install.
Then open Chat under Plugins in the sidebar.
This plugin registers no provider, asks for no login, and stores no credentials. It reuses whatever magpie itself is already signed in to; every request goes out through magpie's gateway, so usage and failover match the agent exactly.
What it does store is only yours, all under magpie's config directory:
~/.config/magpie/magpie-chat/
├── settings.json model, effort, system prompt, ticked instructions/skills/MCP, layout
└── conversations/ one JSON per conversation
magpie's plugin pages run in a sandbox="allow-scripts" frame: no network, no nested frames, no storage. The only way out is to navigate itself somewhere else. So this plugin is two layers:
page/index.html is a small launcher hosted by magpie; it reads the window theme and location.replaces to the local service.index.mjs is loaded as an OpenCode v1 plugin by magpie's Bun host, and starts that local service (127.0.0.1:3477).lib/server.mjs serves /api/* and forwards chat requests to magpie's gateway; chat/ is the real UI.The service binds loopback only and refuses any request whose Origin is neither opaque null (the sandboxed frame) nor its own, so no web page can borrow it to spend your quota.
index.mjs 插件入口(启动本地服务) / plugin entry, starts the local service
page/ 带路页,magpie 托管 / launcher page, hosted by magpie
chat/ 对话界面 / the chat UI (index.html, app.js, style.css, vendor/katex)
lib/server.mjs HTTP 路由、SSE、设置与对话持久化 / routes, SSE, settings, persistence
lib/gateway.mjs magpie 网关客户端 / gateway client
lib/library.mjs 读取指令、技能、MCP 与供应商图标 / reads instructions, skills, MCP, icons
lib/mcp.mjs MCP 客户端(Streamable HTTP + stdio)/ MCP client
assets/icon.svg 插件图标 / plugin icon
无需构建,仓库里的文件就是 magpie 加载的文件。 No build step: the files in this repo are exactly what magpie loads.
MIT
Please log in to share your review and rating for this MCP.
Explore related MCPs that share similar capabilities and solve comparable challenges
by modelcontextprotocol
An MCP server implementation that provides a tool for dynamic and reflective problem-solving through a structured thinking process.
by zylon-ai
Provides an open-source API layer that enables local OpenAI‑compatible models to be used for production AI applications, offering standardized message handling, document ingestion, retrieval‑augmented generation, tool integration, and MCP connectivity.
by danny-avila
Provides a self‑hosted ChatGPT‑style interface supporting numerous AI models, agents, code interpreter, image generation, multimodal interactions, and secure multi‑user authentication.
by block
Automates engineering tasks on local machines, executing code, building projects, debugging, orchestrating workflows, and interacting with external APIs using any LLM.
by RooCodeInc
Provides an autonomous AI coding partner inside the editor that can understand natural language, manipulate files, run commands, browse the web, and be customized via modes and instructions.
by pydantic
A Python framework that enables seamless integration of Pydantic validation with large language models, providing type‑safe agent construction, dependency injection, and structured output handling.
by mcp-use
A Python SDK that simplifies interaction with MCP servers and enables developers to create custom agents with tool‑calling capabilities.
by lastmile-ai
Build effective agents using Model Context Protocol and simple, composable workflow patterns.
by Klavis-AI
Provides production‑ready MCP servers and a hosted service for integrating AI applications with over 50 third‑party services via standardized APIs, OAuth, and easy Docker or hosted deployment.