一页人间 MCP 接入介绍

机器可读版本：/.well-known/mcp.json · 产品事实摘要：/llms.txt · 站点地图：/sitemap.xml

什么是 MCP

MCP（Model Context Protocol）是连接 AI 助手与外部工具、数据源的标准协议。本助手经 MCP 端点自动发现工具，并在对话中按需调用。

端点与鉴权

端点：POST https://yiyerenjian.cn/mcp（Streamable HTTP、无状态，仅支持 POST；GET 返回 404）。所有 JSON-RPC 方法都要求 Authorization: Bearer <token>，未带或无效 token 返回 401。 token 可以是登录 session，也可以是应用内「设置 → API Keys」创建的用户级 API Key（ck_live_ 前缀）。

curl -X POST https://yiyerenjian.cn/mcp \
  -H 'Authorization: Bearer <你的 session token 或 ck_live_ 开头的 API Key>' \
  -H 'Content-Type: application/json' \
  -H 'Accept: application/json, text/event-stream' \
  -d '{"jsonrpc":"2.0","id":1,"method":"tools/list","params":{}}'

当前可用工具（15）

chat Chat — Text chat completion.
generate_image Generate image — Generate image(s) from a text prompt.
edit_image Edit image — Edit a source image_url (e.g. generate_image output). Optional PNG mask: transparent = edit area, must match source dimensions.
remove_watermark Remove watermark — Remove a watermark/logo/stamp and inpaint the area. Use this (NOT edit_image) when the user asks to remove a watermark. Caller MUST supply the removal prompt.
restore_photo Restore photo — Restore/repair an old or damaged photo (NOT edit_image). Caller MUST supply the restore prompt.
ecommerce_image Ecommerce image — Generate one ecommerce image from a source product photo. Requires a reference image (image_refs/image_url); for text-only generation (no product photo) use generate_image instead. Use this (NOT edit_image) for a single ecommerce/product image; for a full set use the ecommerce-image-set skill. Caller MUST supply prompt + business.
text_to_video Text to video — Generate a video from a text prompt (async). Returns video URLs.
image_to_video Image to video — Generate a video from a source image plus a motion prompt (async). Returns video URLs.
reference_to_video Reference to video — Replace the person in a source video with the person from a photo (async). Returns video URLs.
video_to_video Video to video — Edit a source video with a text instruction, optionally bringing in a reference image (async). Returns video URLs.
generate_audio Generate audio — Generate audio (music or speech) from a text prompt. May be async; returns audio URLs.
generate_speech Generate speech — Text-to-speech: synthesize speech from text. May be async; returns audio URLs.
web_search Web search — Search the web and answer with source citations [{ url, title? }].
list_skills List skills — List installed assistant skills (name + when-to-use). No upstream call.
get_skill Get skill — Fetch one skill by name: its playbook (description/clarify/steps/acceptance). Use when a multi-step goal matches a name from list_skills.

如何接入新 MCP

服务端注册新的 MCP 服务或工具后，重新进入助手页即可自动发现；要把本站能力接入 Cursor 等外部 Agent，请到「设置 → API Keys」生成密钥，并按下面的配置填写（也可直接使用创建密钥时返回的 mcpServers 配置）：

{
  "mcpServers": {
    "pixpix": {
      "url": "https://yiyerenjian.cn/mcp",
      "headers": { "Authorization": "Bearer <ck_live_你的密钥>" }
    }
  }
}

常见问题

一页人间 MCP 端点地址是什么？

端点为 POST https://yiyerenjian.cn/mcp（Streamable HTTP、无状态，仅支持 POST；GET 返回 404）。未带 Bearer token 时返回 401。

MCP 怎么鉴权？

请求头 Authorization: Bearer <token>。token 可以是登录后的 session，也可以是在应用内「设置 → API Keys」创建的用户级 API Key（ck_live_ 前缀）；未带或无效 token 返回 401。

MCP 提供哪些工具？

共 15 个：`chat`、`generate_image`、`edit_image`、`remove_watermark`、`restore_photo`、`ecommerce_image`、`text_to_video`、`image_to_video`、`reference_to_video`、`video_to_video`、`generate_audio`、`generate_speech`、`web_search`、`list_skills`、`get_skill`；一句话能力说明见接入介绍页与 llms.txt。

怎么把一页人间接入 Cursor 等外部 Agent？

登录应用 → 「设置 → API Keys」→ 创建密钥，把响应里的 mcpServers 配置粘进客户端配置文件（url 改为 https://yiyerenjian.cn/mcp），重启客户端即可。

联系我们

问题反馈与接入咨询：[email protected]