OpenSpace/examples/parallel-search/README.md

Parallel web search and page fetching

Use Parallel Search MCP through OpenSpace's existing MCP provider for free web search and page fetching, with no account or API key. Anonymous searches use Fast mode. The free endpoint is rate limited and intended for exploration and light use.

Run the example

From the repository root, with Python 3.12 or newer:

python -m venv .venv-parallel
source .venv-parallel/bin/activate
cd examples/parallel-search
python -m pip install -r requirements.txt
OPENSPACE_SKIP_DOTENV=1 python search.py

On Windows, activate with .venv-parallel\Scripts\activate and set OPENSPACE_SKIP_DOTENV=1 in your shell before running python search.py. The example constrains the MCP SDK to 1.x, which the current OpenSpace connectors require.

The script loads config_mcp.json, discovers the live tools through MCPProvider, calls web_search and web_fetch, prints their results, and closes its sessions. It makes direct tool calls without invoking an LLM or requiring LLM credentials. Customize the search and the page to fetch:

OPENSPACE_SKIP_DOTENV=1 python search.py \
  --query "Python asyncio TaskGroup cancellation" \
  --url https://docs.python.org/3/library/asyncio-task.html

Add the server to your OpenSpace configuration

Merge the parallel entry from this example's mcpServers into your openspace/config/config_mcp.json, retaining any existing entries. OpenSpace's HTTP connector negotiates Streamable HTTP with https://search.parallel.ai/mcp; the entry includes an OpenSpace User-Agent and no authentication header. The example does not change the built-in web_search provider. Parallel's web_search and web_fetch are external MCP tools.