Skip to content

Enterprise agent platform for building, deploying, and operating AI teammates.

2.6 KB
Raw Blame

Parallel Search MCP

Parallel Search MCP provides free, keyless web search and page excerpts at https://search.parallel.ai/mcp. It can be connected as a custom MCP server. The connection is opt-in and does not change Xagent's built-in web search provider selection or saved credentials. Free-tier requests are subject to the service's rate limits.

Run the example

From a source checkout with Xagent's core dependencies installed (pip install -e .), run from the repository root:

PYTHONPATH=src python scripts/parallel_search_example.py \
  --config docs/examples/parallel-search/mcp_server.yaml \
  --query "Python asyncio TaskGroup documentation" \
  --fetch-url https://docs.python.org/3/library/asyncio-task.html

The YAML uses Xagent's MCPServerConfig format. The example validates it, converts it to an effective connection, and discovers tools with load_mcp_tools_as_agent_tools. It calls the resulting agent tools for web_search and, when --fetch-url is supplied, web_fetch. The JSON output contains source URLs and excerpts in Xagent's normal MCP result shape. A fresh session identifier is reused across both calls. No LLM credentials are needed to run this tool example.

This standalone example does not initialize Xagent's database or modify saved connectors. It uses Streamable HTTP, no authorization header, and the xagent/parallel-search-example User-Agent from the YAML. Discovery, execution, error handling, cancellation, and transport timeouts use the existing MCP runtime. To fetch a source discovered by search, run the example again with that URL as --fetch-url.

Connect in Xagent

In the Connectors page, add a Custom MCP connection with these values:

Setting Value
Name parallel_search
Transport streamable_http
URL https://search.parallel.ai/mcp
Authentication None
Header User-Agent: xagent/parallel-search-example

Load the server's tools and select the connector for the agent that needs it. Its search and fetch tools appear with the MCP server prefix, separately from the built-in search tools. Removing the connector from that agent removes its access to these tools. This does not select Parallel as the built-in search provider.

Validate the example

With the core dependencies installed, the local protocol test starts a loopback MCP server and exercises the example command, tool loading, search/fetch calls, request headers, session reuse, and a server error:

PYTHONPATH=src python -m unittest discover -s tests/scripts -p test_parallel_search_example.py