vision4nx-kb-mcp

vision4nx-kb-mcp

MCP server that exposes the Vision 4 NX knowledge base to any MCP client, enabling listing, querying, and reading of knowledge base contents via RAG vector search.

Category
访问服务器

README

vision4nx-kb-mcp

MCP server (streamable HTTP) that exposes the Vision 4 NX knowledge base to any MCP client (Claude Code, Claude Desktop, and other MCP-capable clients).

It is a thin wrapper over the Vision 4 NX REST API — embedding, hybrid search and reranking all happen server-side, so this server needs no vector-DB access and no embedding model.

Tools

Tool What it does Vision 4 NX endpoint
list_knowledge_bases() List accessible KBs (id, name, description) GET /api/v1/knowledge/ (paginated)
query_knowledge_base(query, knowledge_base_ids, k=5) RAG vector search; returns relevant chunks with source file info POST /api/v1/retrieval/query/collection
list_knowledge_base_files(knowledge_base_id) Files inside one KB GET /api/v1/knowledge/{id}
read_file_content(file_id, max_chars=50000) Full extracted text of a file GET /api/v1/files/{id}/data/content

Setup

1. Get your access credentials

The VISION4NX_URL and VISION4NX_API_KEY for the Vision 4 NX instance are issued on request. Contact the Inteliscience team at info@inteliscience.net to obtain them.

The MCP server acts with this single service key — all clients share the KB permissions granted to it.

If tools return 403 or 404 errors, the access token may be missing permissions for a knowledge base — contact Inteliscience to have it adjusted.

2. Configure

cp .env.example .env   # set VISION4NX_URL and VISION4NX_API_KEY (from Inteliscience)

3. Run

Docker (recommended):

docker compose up -d --build

Plain Python (>=3.11):

python3.12 -m venv .venv
.venv/bin/pip install -e .
.venv/bin/vision4nx-kb-mcp
# or: .venv/bin/uvicorn vision4nx_kb_mcp.server:app --host 0.0.0.0 --port 8600

Health check: curl http://localhost:8600/health MCP endpoint: http://localhost:8600/mcp

Connecting clients

This server speaks streamable HTTP at http://localhost:8600/mcp. Clients that support HTTP/remote MCP connect to that URL directly; stdio-only clients need the mcp-remote bridge (shown below). No auth token is required — client auth is not enforced by this server.

Claude Code:

claude mcp add --transport http vision4nx-kb http://localhost:8600/mcp

Claude Desktop: two ways, depending on your version.

  • Custom connector (if available): Settings → Connectors → Add custom connector, name it Vision 4 NX KB, URL http://localhost:8600/mcp, save and enable.
  • Config file (works everywhere, needs Node.js): the desktop config only launches stdio commands, so bridge the HTTP endpoint with mcp-remote. Edit ~/Library/Application Support/Claude/claude_desktop_config.json (macOS) or %APPDATA%\Claude\claude_desktop_config.json (Windows):
    {
      "mcpServers": {
        "vision4nx-kb": {
          "command": "npx",
          "args": ["-y", "mcp-remote", "http://localhost:8600/mcp"]
        }
      }
    }
    
    Then fully quit and reopen Claude Desktop (Cmd+Q / quit from the tray — closing the window is not enough).

Vision 4 NX itself (use the KB tools from chats): Admin Settings → External Tools → add a tool server of type MCP with URL http://localhost:8600/mcp (from inside the compose stack: the container-network URL).

Any other MCP client: point it at http://localhost:8600/mcp with transport Streamable HTTP. If the client only supports stdio, wrap it the same way Claude Desktop's config file does: npx -y mcp-remote http://localhost:8600/mcp.

MCP Inspector (debugging):

npx @modelcontextprotocol/inspector
# transport: "Streamable HTTP", URL: http://localhost:8600/mcp

Deploying next to the Vision 4 NX stack

Uncomment the networks block in docker-compose.yaml, verify the network name (docker network ls), and point VISION4NX_URL at the app container (http://vision4nx:8080). Optionally drop the published port and proxy /mcp through the existing nginx instead.

Environment variables

Var Default Purpose
VISION4NX_URL — (required) Base URL of the Vision 4 NX instance (issued by Inteliscience)
VISION4NX_API_KEY — (required) Service access token (issued by Inteliscience — info@inteliscience.net)
MCP_HOST / MCP_PORT 0.0.0.0 / 8600 Server bind
MCP_AUTH_TOKEN empty Reserved for future client auth — currently ignored
LOG_LEVEL INFO Logging level

Tests

Opt-in e2e test against a running server:

pip install -e '.[dev]'
MCP_SERVER_URL=http://localhost:8600/mcp pytest tests/test_e2e.py

Compatibility

Built against the Vision 4 NX backend, which paginates GET /api/v1/knowledge/ and pins mcp==1.26.0. The paginated-list handling falls back to an unpaginated response shape automatically.

推荐服务器

Baidu Map

Baidu Map

百度地图核心API现已全面兼容MCP协议,是国内首家兼容MCP协议的地图服务商。

官方
精选
JavaScript
Playwright MCP Server

Playwright MCP Server

一个模型上下文协议服务器,它使大型语言模型能够通过结构化的可访问性快照与网页进行交互,而无需视觉模型或屏幕截图。

官方
精选
TypeScript
Magic Component Platform (MCP)

Magic Component Platform (MCP)

一个由人工智能驱动的工具,可以从自然语言描述生成现代化的用户界面组件,并与流行的集成开发环境(IDE)集成,从而简化用户界面开发流程。

官方
精选
本地
TypeScript
Audiense Insights MCP Server

Audiense Insights MCP Server

通过模型上下文协议启用与 Audiense Insights 账户的交互,从而促进营销洞察和受众数据的提取和分析,包括人口统计信息、行为和影响者互动。

官方
精选
本地
TypeScript
VeyraX

VeyraX

一个单一的 MCP 工具,连接你所有喜爱的工具:Gmail、日历以及其他 40 多个工具。

官方
精选
本地
graphlit-mcp-server

graphlit-mcp-server

模型上下文协议 (MCP) 服务器实现了 MCP 客户端与 Graphlit 服务之间的集成。 除了网络爬取之外,还可以将任何内容(从 Slack 到 Gmail 再到播客订阅源)导入到 Graphlit 项目中,然后从 MCP 客户端检索相关内容。

官方
精选
TypeScript
Kagi MCP Server

Kagi MCP Server

一个 MCP 服务器,集成了 Kagi 搜索功能和 Claude AI,使 Claude 能够在回答需要最新信息的问题时执行实时网络搜索。

官方
精选
Python
e2b-mcp-server

e2b-mcp-server

使用 MCP 通过 e2b 运行代码。

官方
精选
Neon MCP Server

Neon MCP Server

用于与 Neon 管理 API 和数据库交互的 MCP 服务器

官方
精选
Exa MCP Server

Exa MCP Server

模型上下文协议(MCP)服务器允许像 Claude 这样的 AI 助手使用 Exa AI 搜索 API 进行网络搜索。这种设置允许 AI 模型以安全和受控的方式获取实时的网络信息。

官方
精选