You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
[docs] Add an optional Parallel Search MCP client example #2131
A developer using Commonly from an MCP client can read pod context and collaborate, but the setup guide does not show how to add web research alongside those tools.
Proposed Solution
Add an opt-in runnable client configuration example pairing @commonlyai/mcp with Parallel Search MCP at https://search.parallel.ai/mcp. It provides free web search in Fast mode and page fetching without a Parallel account or API key, for exploration and light use.
The example would extend the existing MCP client setup, keep Commonly authentication and defaults unchanged, and cover a supported client that can load both stdio and HTTP servers. This is for the user-driven MCP connection path; Commonly's daemon and wrapper adapters have separate restrictions on remote servers.
Alternatives Considered
Users can configure another research server themselves. A native provider or daemon change would broaden the scope; a focused example can use the supported client setup directly.
Area
Developer experience / SDK
Use Case
Read a pod's context, search for current sources, fetch a relevant page, and share a sourced answer back to the pod. Before a PR, verify installation, tool discovery and dispatch through the documented client path, including anonymous search and fetching.
Thanks, George, and sorry this sat for three weeks with no reply.
It's merged: #3048 adds the section to docs/MCP_INTEGRATION.md. It covers Claude Code and Cursor, and walks through your use case: read the pod, search, fetch the page, post a sourced answer back.
We pointed it at https://mcp.parallel.ai/mcp instead of search.parallel.ai, since your docs now recommend that one for new installs. The doc notes that the standalone server still runs. Anonymous web_search and web_fetch both worked when we tested them with the MCP SDK client. We didn't run Cursor itself, so if that config block looks off to you, let us know.
Now that Parallel is in our docs, would your team be up for doing more together? A design partnership, or some co-marketing around this integration. Happy to take that to email or a call if it's easier.
Activity
lilyshen0722 commentedon Oct 31, 2026
Thanks, George, and sorry this sat for three weeks with no reply.
It's merged: #3048 adds the section to docs/MCP_INTEGRATION.md. It covers Claude Code and Cursor, and walks through your use case: read the pod, search, fetch the page, post a sourced answer back.
We pointed it at https://mcp.parallel.ai/mcp instead of search.parallel.ai, since your docs now recommend that one for new installs. The doc notes that the standalone server still runs. Anonymous web_search and web_fetch both worked when we tested them with the MCP SDK client. We didn't run Cursor itself, so if that config block looks off to you, let us know.
Now that Parallel is in our docs, would your team be up for doing more together? A design partnership, or some co-marketing around this integration. Happy to take that to email or a call if it's easier.