ApeMind homeRAG ArchitectureSolutionsEnterpriseCompareSecurityKnowledge GraphAgent RSIDocs
中文版

APEMIND / PROTOCOLS, IM CHANNELS & CONNECTORS

Connect ApeMind RAG to Cursor, Claude Code, Feishu, Slack, databases & APIs

An enterprise knowledge base only creates leverage when teams can query it inside the tools they already use. ApeMind exposes open developer protocols (MCP, OpenAI-compatible REST, A2A 1.0, declarative API, CLI), publishes agents to five enterprise IM channels and website embeds, and syncs inbound knowledge from Feishu, websites, email, and relational databases.

Verified integration matrix

Every protocol, connector, and channel listed below is verified against ApeMind documentation and runtime code, alongside its explicit access and safety boundaries.

LayerProtocol / ChannelVerified ApeMind capabilityExplicit scope & boundaryDoc
Developer ProtocolBuilt-in MCP Server (/mcp/)Streamable HTTP MCP server exposing 20 built-in tools for knowledge search, passage inspection, graph traversal, and agent interaction—connect directly from Cursor, Claude Code, or custom MCP clients.Authenticated via workspace-bound API keys; tool visibility and collection access stay strictly within the key’s organization and workspace scope.MCP Server
Developer ProtocolOpenAI-Compatible APIPOST /v1/chat/completions endpoint supporting Python and TypeScript OpenAI SDKs, SSE streaming, and multi-turn conversation against configured ApeMind agents.Scoped to agent-level API keys and supported OpenAI Chat Completions fields; non-standard provider extensions are documented in the reference.OpenAI API
Developer ProtocolA2A 1.0 ProtocolPrivate Agent Card, JSON-RPC, and SSE streaming at /v1/a2a/bots/{bot_id} to integrate ApeMind agents into external multi-agent orchestrators.A2A v1 supports single-turn asynchronous tasks (no multi-turn follow-up or push notifications inside a single task) and requires API-key authentication.A2A 1.0
Developer ProtocolDeclarative Config & CLIManage agents, templates, and collections via metadata + spec declarative APIs with dry-run previews, plus the standalone apemind CLI for automation and evaluation loops.Installing the standalone CLI connects to an existing ApeMind server and does not deploy a local backend.Declarative API
IM ChannelsFeishu / Lark, DingTalk, WeCom, Slack, TelegramPublish one agent to five workplace messengers simultaneously for DM Q&A and group @mentions, with thread context continuity and per-channel latency/traffic monitoring.Agents holding organization-admin privileges cannot be bound to IM channels; write actions requiring interactive confirmation are automatically blocked in IM chats.IM channels
Web & SharingPublic Links, Website Bubble & Agent API KeysShare agents with org teammates, issue login-free public links, embed a floating website chat bubble, or mint agent-scoped API keys with three-tier access boundaries.Public share links serve sanitized frozen citation snapshots so internal collections and revoked documents are not exposed.Agent access
Inbound SyncFeishu Wiki/Drive, Websites & IMAP EmailScheduled daily, weekly, or monthly incremental collection sync from Feishu knowledge spaces and Drive, full-site web crawls, and IMAP mailboxes, with reviewable source-deleted cleanup.Sync credentials are managed at organization scope; deletion of removed upstream files requires operator review rather than silent permanent wipe.Data sources
External Tools & DBsPostgreSQL, MySQL, GitHub, Remote MCP & HTTP/OpenAPIOrganization-managed Connections for read-only SQL database knowledge bases, GitHub, remote Streamable HTTP MCP servers, and HTTP/OpenAPI 3.0–3.1 APIs with per-operation approval.Database connections enforce read-only AST + read-only transactions (max 200 rows, 256 KiB, 30s) and still require a dedicated read-only DB user.Connections

Frequently asked questions about ApeMind integrations

Can I connect ApeMind knowledge bases to Cursor or Claude Code via MCP?

Yes. ApeMind ships a built-in Streamable HTTP MCP Server at /mcp/ with 20 tools covering knowledge retrieval, passage readback, and graph exploration. Configure the server URL and a workspace-scoped API key in Cursor, Claude Code, or any MCP-compatible client.

Which enterprise IM platforms are supported out of the box?

ApeMind supports five IM channels out of the box: Feishu / Lark (long connection), DingTalk (Stream mode), WeCom (callback API), Slack (Socket Mode), and Telegram (BotFather). A single agent can serve multiple channels at once with per-channel monitoring.

Does ApeMind require a public callback URL for Feishu, DingTalk, or Slack bots?

No for Feishu / Lark (long connection), DingTalk (Stream mode), and Slack (Socket Mode)—these channels work behind private corporate firewalls with outbound connections only. WeCom requires its callback API verification.

Can existing applications call ApeMind using the standard OpenAI SDK?

Yes. Point the Python or TypeScript OpenAI SDK base URL to ApeMind’s OpenAI-compatible endpoint (POST /v1/chat/completions) with an agent-scoped API key to stream responses and preserve multi-turn conversations.

Continue reading

  • Enterprise RAG architecture: PDF parsing, Jev chunking & RRF
  • Compare ApeMind vs Dify, RAGFlow, FastGPT & other RAG platforms
  • Security architecture, storage topology & credential boundaries
  • Download ApeMind Desktop & standalone CLI
  • Developer integration guide (Chinese)

Wire enterprise knowledge into your existing workflow

Start with a workspace API key for MCP or OpenAI SDK integration, or publish an agent directly to your team’s IM group.

Open ApeMind workspace