mcpField
C
canvas-mcp

canvas-mcp

Remote
agent.avatargaia.topAI & Agents0Updated 3w ago

About canvas-mcp

@avatargaia/canvas-mcp is a stateless stdio bridge for connecting compatible MCP clients to TeamAgent’s streamable HTTP endpoints. It forwards requests and supports Canvas tools for staff, actions, data, and knowledge bases; access varies by key and endpoint.

Key features

  1. 01stdio-to-HTTP request forwarding
  2. 02Canvas and AICare endpoint selection
  3. 03Staff setup and configuration tools
  4. 04Knowledge base publishing and assignment
  5. 05MCP tool annotations and error forwarding

Use cases

  1. 01Connect a stdio-only client to TeamAgent
  2. 02Set up staff and assign knowledge bases
  3. 03Publish and manage wiki content
  4. 04Inspect upstream errors in the client

MCP support

It bridges stdio MCP clients to TeamAgent MCP endpoints using streamable HTTP. The upstream endpoints expose Canvas and AICare tools; available tools depend on endpoint and authentication.

Configuration

Remote servers need no local install. Pick your client, then copy the config or install in one click.

Endpointhttps://agent.avatargaia.top/api/mcp/canvas
Copy install command
Claude CodeTerminal
Cursor.cursor/mcp.json
VS Code.vscode/mcp.json
Claude Desktopclaude_desktop_config.json
Windsurf~/.codeium/windsurf/mcp_config.json

Run commands in a terminal; paste JSON into the client’s MCP config and restart. Servers that need sign-in will prompt on first connect.

All endpoints
Streamable HTTPhttps://agent.avatargaia.top/api/mcp/canvasAuth required
Packages
npm@avatargaia/canvas-mcp0.1.4

Tools

18

This tool takes no parameters.

Details

From the project README

@avatargaia/canvas-mcp

stdio 马甲:让只支持 stdio 的 MCP 客户端(Claude Desktop / Cursor / Cline / 各类本地 Agent)接入 TeamAgent 的 MCP 服务器(streamable-http)。 纯转调,无状态、不缓存、不改写请求。

两个门,同一把 Key:

门端点工具鉴权
Canvashttps://agent.avatargaia.top/api/mcp/canvas带 Key 97 个;匿名沙箱 18 个免费工具可不带 Key:匿名沙箱开放免费工具;带 dk_/ta_ 解锁全部
AICarehttps://agent.avatargaia.top/api/mcp/aicare7 个(AI 健康检测 13 类:舌苔/面部/指甲…)必须带 Key,按次计费

自动安装说明(给 Agent 读):llms-install.md | 隐私政策与支持:https://agent.avatargaia.top/privacy

用

CANVAS_MCP_URL=https://agent.avatargaia.top/api/mcp/canvas \
CANVAS_MCP_TOKEN=ta_xxxxxxxx \
npx -y @avatargaia/canvas-mcp

客户端配置(Claude Desktop claude_desktop_config.json 为例):

{
  "mcpServers": {
    "teamagent-canvas": {
      "command": "npx",
      "args": ["-y", "@avatargaia/canvas-mcp", "--token", "dk_xxxxxxxx"]
    }
  }
}

环境变量

变量必填说明
CANVAS_MCP_TOKEN否dk_(开发者 Key)或 ta_(平台令牌);不带 = 匿名沙箱,只能用免费工具
CANVAS_MCP_URL否默认 Canvas 门;AICare 门填 https://agent.avatargaia.top/api/mcp/aicare;自建部署时改这里
CANVAS_MCP_KEY否CANVAS_MCP_TOKEN 的别名

给数字员工喂知识(v0.4.0 起)

岗位可以配知识库:访客问到时,数字员工先查库再答,并把依据的那一页显示在画布上;库里没有的,它会直说没有。

工具作用
list_kbs我名下的知识库 + 每个岗位配了哪些库
read_kb库的页面清单,或某一页原文
publish_kb发布一套 wiki(整库替换,可顺手配给岗位);checked 不为 true 不发
set_staff_kbs设定岗位用哪些库
hire_staff按模板开岗,可带 kbs

三步:整理(一套 Markdown,wiki/index.md 总目录 + wiki/<主题>/<文章>.md,一个概念一篇,数字日期照原文写)→ 发布(publish_kb,库名 + 全部页面 + checked: true,同名再发是整库替换)→ 配给岗位(set_staff_kbs,或发布时带 slug,或 hire_staff 直接带 kbs)。

⚠️ 发布等于公开:岗位是匿名访客都能聊的,库里写了什么访客就能问出什么。个人隐私、客户资料、令牌、内部运维内容不要放。上限:200 个文件 / 单个 64KB / 合计 2MB。

行为

  • stdin/stdout 逐行 JSON-RPC 2.0;通知(无 id)不回包
  • 上游 HTTP 错误原样透传成 JSON-RPC error(401: Missing Authorization header / 402: … / 429: …),客户端能看见真正原因
  • 上游返回 SSE 或 JSON 都能解析(取最后一条 data:)
  • stdin 关闭后等所有在途请求落地再退出

验证记录(2026-10-03)

initialize   ✅ serverInfo teamagent-canvas 0.4.0
tools/list   ✅ 91 个工具
publish_kb   ✅ checked:false 被拦下;checked:true 发布成功
hire_staff   ✅ 开岗并同时配知识库,返回里带 kbs
set_staff_kbs ✅ 正向成功;跨账号岗位返回 404(越权被拒)
匿名调用      ✅ {"error":{"code":-32000,"message":"401: Missing Authorization header"}}

(早期 0.3.0 的 stdio 马甲验证:initialize / tools/list / tools/call / 无令牌 401 / mcporter 接入,均通过。)

拿 Key(自助,30 秒)

外部 / 本地 Agent 无需 TeamAgent 账号:

curl -X POST https://agent.avatargaia.top/api/dev/register \
  -H "Content-Type: application/json" \
  -d '{"name":"my-agent","email":"[email protected]"}'
# → 返回 dk_ Key(只显示一次)+ 20 credits + 接入信息

工具注解

每个工具都带 MCP annotations(readOnlyHint / destructiveHint / idempotentHint / openWorldHint),付费工具在描述里写明 credit 数,客户端可据此做二次确认。

收录

Comments

Markdown supported

New to MCP?

From what the protocol is to installing your first server, these answers cover most first questions.

An open protocol that defines how AI applications connect to external tools and data in a uniform way. Clients talk to the model; MCP servers provide tools, resources and prompt templates. They exchange standard messages, so one server can be reused by many clients.