引力AI

Developer guide

引力AI MCP: a human network for AI Agents

引力AI MCP connects compatible AI Agents to an authorized human network. The hosted endpoint supports discovery around needs, interests, professional experience and skills, plus publication and private-chat workflows controlled by user confirmation.

引力AI·Updated 中文

Connection and authentication

Endpoint: https://api.fitmeet.cn/api/v1/mcp. Transport: Streamable HTTP. Authentication: per-user OAuth authorization code flow with PKCE S256. Dynamic client registration is supported.

Use the client’s own OAuth flow. Do not configure a shared access token or paste credentials into a conversation.

What the tools do

The MCP 2.3 service catalog provides 23 tools across seven scopes: group metadata, owner notifications, connection items and feedback; profile and people discovery; publication sources, preview and confirmation; conversation and message reading; chat and message preview and confirmation; plus inquiry and group prepare/confirm actions. Tool availability depends on the current authorization scopes and live tools/list response.

The scopes are profile:read, people:search, hall:publish, messages:read, messages:write, social:read and social:write. Inquiry, group creation and joining a public group use a short-lived prepare → confirm receipt. Group scheduling, private-group invitations and feedback edits remain in the linked 引力AI pages. Retrieve current tool schemas from the service rather than inventing parameters.

Authorization is not action confirmation

OAuth grants a capability scope. It does not approve a particular publication, message, inquiry or group action. Show the exact preview and obtain the user’s explicit confirmation before using a confirm tool. Opening a conversation creates an empty chat; it does not send a message.

External discovery is controlled by the profile owner. Search results retain uncertainty and do not guarantee identity, availability or outcomes. The MCP does not provide payment, booking or administrator capabilities.

Check the exact client

The public MCP and Skill package and the DeepSeek Harness plugin are available in their repositories. Check installation requirements and current verification status. No claim of universal client compatibility or marketplace approval is made.

Source and corrections

Prepared with AI assistance from 引力 product documentation and engineering records. Examples are illustrative; client-specific acceptance and real-user outcomes are stated separately.

Publisher and company information (Chinese) →

Report an inaccuracy (Chinese support) →