Bing Web, News & Image Search MCP Server

Overview

This MCP (Model Context Protocol) server connects AI assistants to Microsoft Bing Search APIs, enabling programmatic web, news, and image searches from within MCP-compatible clients (for example Claude Desktop or Cursor). The server exposes a set of tools that let a model request search results and media metadata, letting the assistant augment responses with up-to-date information, citations, and images.

Built as a lightweight Python service, the server handles request translation, authentication with the Bing Search API, rate limiting to protect your quota, and consistent error handling so downstream agents can rely on predictable results. The project repository is available at https://github.com/leehanchung/bing-search-mcp.

Features

• Web search for general-purpose information and site links
• News search for recent events and topical coverage
• Image search for thumbnails, metadata, and preview URLs
• Built-in rate limiting to reduce accidental API overuse
• Structured error reporting for graceful fallback in clients
• Minimal configuration and MCP-compatible entrypoint for easy integration

Requirements

• Python 3.10 or newer
• A Microsoft Bing Search API key (from Azure)
• An MCP-compatible client or assistant that can invoke MCP servers

Installation / Configuration

1. Clone the repository and install in a virtual environment:

git clone https://github.com/leehanchung/bing-search-mcp.git
cd bing-search-mcp
python -m venv .venv
source .venv/bin/activate   # Windows: .venv\Scripts\activate
pip install -e .

2. Set required environment variables (Linux / macOS):

export BING_API_KEY="your-bing-api-key"
export BING_API_URL="https://api.bing.microsoft.com/"  # Optional: default used if not set

Windows cmd:

set BING_API_KEY=your-bing-api-key
set BING_API_URL=https://api.bing.microsoft.com/

3. Run the MCP server (example using uvx entrypoint installed by the package):

uvx bing-search-mcp

To run under a process manager or system service, start the same command in the desired environment.

Available Tools

The server exposes three primary tools (endpoints) via MCP. Each tool returns structured search results suitable for downstream consumption by an assistant.

Example function signatures (logical API):

bing_web_search(query: str, count: int = 10, offset: int = 0, market: str = "en-US")
bing_news_search(query: str, count: int = 10, market: str = "en-US", freshness: str = "Day")
bing_image_search(query: str, count: int = 10, market: str = "en-US")

Returned data includes title, snippet/description, url, and metadata fields (e.g., provider, datePublished). Image results include thumbnail and content URLs, sizes, and image type when available.

Use Cases

• Fact-checking and real-time references: an assistant can call bing_news_search to verify recent claims with links and publication dates.
• Enriching answers with sources: use bing_web_search to collect top authoritative links and present citations alongside generated summaries.
• Visual assistance and content discovery: use bing_image_search to provide images or thumbnails for UI previews, documentation, or creative workflows.

Concrete example (assistant flow):

1. User asks “What are the latest developments on electric vehicle batteries?”
2. Assistant calls bing_news_search(“electric vehicle battery developments”, count=5, freshness=“Week”)
3. Assistant summarizes the top 3 articles, includes publication dates, and attaches source links.

Integration with MCP Clients

For Claude Desktop, add an MCP server entry to your configuration:

{
  "mcpServers": {
    "bing-search": {
      "command": "uvx",
      "args": [ "/path/to/your/bing-search-mcp" ],
      "env": { "BING_API_KEY": "your-bing-api-key" }
    }
  }
}

Other MCP-capable clients should accept a similar launch specification and environment mapping.

Rate Limiting and Error Handling

• The server applies request rate limiting to avoid hitting Bing API quotas. Exceeding limits returns clear MCP error payloads so the assistant can retry later or use alternative sources.
• Errors from the Bing API are normalized into predictable error codes and human-readable messages, including HTTP status, error type, and recommended action.

Getting a Bing API Key

1. Sign in to Microsoft Azure Portal (https://portal.azure.com/).
2. Create a Bing Search v7 (or relevant) resource.
3. Open the resource and copy the key under “Keys and Endpoint”.

Troubleshooting

• 401 Unauthorized: verify BING_API_KEY and that the key is active for Bing Search.
• Empty results: try widening the market or increasing count; check that the query text is appropriate.
• High latency: confirm network connectivity to the Bing endpoint and consider caching frequent queries.

License and Source

This project is MIT-licensed. Source, issues, and contributions are hosted on GitHub: https://github.com/leehanchung/bing-search-mcp.