Yandex Search MCP Server
The Yandex Search MCP Server acts as a powerful set of "eyes" for AI assistants, allowing models like Claude or Cursor to browse the live web. Instead of relying solely on their training data, which might be outdated, this tool gives AI agents the ability to look up real-time information, …
About this Protocol
How to Use
1. Installation
Remote Installation (Recommended)
No local installation is required if using the remote SSE connection via npx.
Local Installation
If you prefer to run the server locally, clone the repository and install dependencies:
git clone https://github.com/yandex/yandex-search-mcp-server.git
cd yandex-search-mcp-server
- For Python:
pip install -r requirements.txt - For Docker:
docker build -t yandex-mcp-server-image:latest .
2. Configuration
To use this server, you must have a Yandex Search API Key (with yc.search-api.execute scope) and a Folder ID (with search-api.editor role).
Claude Desktop
Add this to your claude_desktop_config.json:
{
"mcpServers": {
"yandexSearch": {
"command": "npx",
"args": [
"-y",
"mcp-remote",
"https://d5dj4o5pbnqgca1d546v.cmxivbes.apigw.yandexcloud.net:3000/sse",
"--header", "ApiKey:<your_api_key>",
"--header", "FolderId:<your_folder_id>"
]
}
}
}
Cursor
Add a Custom MCP in Settings with the following configuration:
* Type: sse
* URL: https://d5dj4o5pbnqgca1d546v.cmxivbes.apigw.yandexcloud.net:3000/sse
{
"mcpServers": {
"yandexSearch": {
"type": "sse",
"url": "https://d5dj4o5pbnqgca1d546v.cmxivbes.apigw.yandexcloud.net:3000/sse",
"headers": {
"ApiKey": "<your_api_key>",
"FolderId": "<your_folder_id>"
}
}
}
}
Local Python Configuration (Standard MCP Client)
{
"mcpServers": {
"yandex-search-api": {
"autoApprove": [],
"disabled": false,
"timeout": 60,
"type": "stdio",
"command": "env",
"args": [
"SEARCH_API_KEY=<your_api_key>",
"FOLDER_ID=<your_folder_id>",
"python3",
"/path/to/yandex-search-mcp-server/server.py"
]
}
}
}
3. Available Tools
- ai_search_post: Performs a real-time web search and returns an AI-generated answer based on results using the Yandex Yazeka AI model.
- web_search_post: Performs a real-time web search and returns results with source citations.
4. Example Prompts
Natural Language (Claude/Cursor):
* "use yandexSearch to find information about the best AI Agent frameworks in 2025"
* "Search for the latest news regarding Yandex Cloud updates using yandexSearch."
Direct Tool Input (JSON):
{
"query": "Who won the most recent Formula 1 race in 2025?",
"search_region": "en"
}
Use Cases
Use Case 1: Coding with Post-Cutoff Frameworks and APIs
Problem: Developers often work with libraries or frameworks that have been updated or released after the AI model's training data cutoff. Asking an LLM for help with a "2025 version" of an API usually results in hallucinations or outdated code snippets.
Solution: By integrating the Yandex Search MCP into an IDE like Cursor or VS Code, the AI can perform a real-time search for the latest documentation. It can retrieve the exact syntax for new functions and apply them directly to the user's codebase.
Example: A developer asks Cursor, "Update my project to use the latest Yandex Cloud SDK features released this month." Cursor uses web_search_post to find the new SDK documentation and suggests the correct updated code based on the search results.
Use Case 2: Instant Fact-Checking and Summarization
Problem: Users often need to synthesize information from multiple news sources or verify a recent event, which usually requires opening several browser tabs and manually reading through them to find a consensus.
Solution: The ai_search_post tool uses the Yandex Yazeka AI model to search the web and generate a single, coherent answer based on the most relevant results. This provides a "Perplexity-like" experience directly within Claude or other MCP-compatible clients.
Example: A user asks Claude, "What were the key highlights of the international technology forum held yesterday?" Claude calls the ai_search_post tool, searches the latest news, and provides a summarized bullet-point list of events with the relevant context.
Use Case 3: Competitive Market Research with Cited Sources
Problem: Market analysts need to identify new competitors or localized business trends, but they require the original sources to verify the data for formal reports.
Solution: Using the web_search_post tool, an AI agent can perform deep searches across the web and return not just the information, but the specific source URLs. This allows the user to click through to the primary source for further validation.
Example: An analyst tells their AI assistant, "Find the top 5 emerging fintech startups in the CIS region and provide links to their official websites and recent funding news." The MCP fetches the data and presents a table with the company name, a brief description, and the verified source link for each entry.
Use Case 4: Debugging Environment-Specific Errors
Problem: When encountering a specific error message (e.g., a specific Linux kernel error or a niche cloud deployment failure), the solution is often buried in a recent GitHub issue or a forum post that the AI hasn't been trained on.
Solution: The AI can take the specific error log provided by the user and use the Yandex Search MCP to find troubleshooting steps from recent community discussions, Stack Overflow posts, or GitHub issues.
Example: A user pastes a cryptic deployment error into VS Code. The AI agent automatically selects the web_search_post tool to search for the specific error string, finds a GitHub issue from two days ago with a workaround, and explains the fix to the user.
