Baidu AI Search MCP Server Real-Time Web LLM

Overview

Baidu AI Search MCP Server Real-Time Web LLM integrates Baidu’s web search pipeline with a large language model interface exposed via a Model Context Protocol (MCP) server. It enables LLM responses that reference fresh web content (RAG-style) and exposes a standardized, real-time API for streaming or synchronous consumption. The server is designed to fit into diverse industry scenarios—customer support, research, assistant workflows—where answers must cite or be grounded in current web sources.

By routing search, query rewriting, and LLM composition through a single MCP endpoint, developers get configurable search scope, persona/model controls, and multi-turn query rewriting without building a custom retrieval layer. The MCP server returns both generated text and structured reference metadata (links, snippets, timestamps), making it easier to build defensible, source-aware applications.

Features

• Real-time web referencing: LLM responses include source links and snippets from current web pages.
• Query rewriting: supports time-sensitive and multi-turn rewriting to improve retrieval relevance.
• Persona and model selection: control response style and model backend.
• Search scope controls: filter by modality (text/image), site ranges, and publication date.
• Adjustable number of references: set how many links the LLM should return and cite.
• Standardized MCP API: SSE/streaming support for live assistant responses and compatibility with AppBuilder integrations.

Installation / Configuration

1. Obtain your AppBuilder API Key from the AppBuilder console.
2. Format the authorization token exactly as: Bearer+<AppBuilder API Key> (note the plus sign).
3. Add the MCP server configuration to your client/app configuration:

Example JSON configuration:

{
  "mcpServers": {
    "baidu_ai_search": {
      "url": "https://appbuilder.baidu.com/v2/ai_search/mcp/sse?api_key=Bearer+<YOUR_APPBUILDER_API_KEY>"
    }
  }
}

Example cURL request (streaming SSE endpoint):

curl -N -H "Accept: text/event-stream" \
  "https://appbuilder.baidu.com/v2/ai_search/mcp/sse?api_key=Bearer+<YOUR_APPBUILDER_API_KEY>" \
  -d '{
    "type":"chat.completions.create",
    "model":"qianfan-erniebot",
    "messages":[{"role":"user","content":"What are the latest regulations for data privacy in China?"}],
    "search": {
      "rewrite": {"mode":"time_sensitive"},
      "scope": {"sites":["gov.cn"], "date_from":"2023-01-01"},
      "max_references": 3
    }
  }'

Notes:

• Use the SSE (/mcp/sse) endpoint for streaming updates and reference insertion events.
• For standard request/response flows, consult the AppBuilder MCP client patterns or SDK wrappers.

Available Resources

• GitHub (source & examples): https://github.com/baidubce/app-builder/tree/master/python/mcp_server/ai_search
• Official documentation: https://cloud.baidu.com/doc/AppBuilder/s/zm8pn5cju
• Common parameters:
  • model: backend LLM name (e.g., qianfan/erniebot variants)
  • persona: tone/style presets
  • rewrite.mode: e.g., “time_sensitive”, “multi_turn”
  • scope: {modalities, sites, date_from, date_to}
  • max_references: integer number of links to include

Example parameter table:

Use Cases

• Customer support agent with live evidence:

  • Query: “Has product X received a safety recall?”
  • Behavior: rewrite to include manufacturer and model, search recent press and government notices, return answer with 2–4 cited links and excerpts.

• Regulatory monitoring and compliance:

  • Query: “What new privacy laws in 2024 affect user tracking?”
  • Behavior: use time-sensitive rewrite, restrict scope to government/legal domains, include publication dates for each reference.

• Research assistant for product teams:

  • Query: “Compare recent mobile chips for power efficiency”
  • Behavior: aggregate benchmark pages, cite comparative articles and whitepapers, and summarize differences with links.

• Multi-turn conversational agents:

  • Example flow: User asks a general question, follow-up clarifies constraints; the MCP server rewrites the follow-up in context and re-runs retrieval, returning an updated answer that cites sources.

Getting Started Tips

• Tune max_references and date filters based on your trust and freshness needs.
• Use persona to maintain consistent assistant voice for user-facing applications.
• Monitor SSE events for partial answers and incremental reference updates when building streaming UIs.
• Review the GitHub examples for Python MCP client patterns and to adapt the configuration to your environment.

For implementation specifics and sample code, see the GitHub repository linked above and the official AppBuilder documentation.