Skip to content

WIP: Add Parallel search and fetch tools for WebMCP sites - #51

Draft
georgeatparallel wants to merge 10 commits into
parallel-web:mainfrom
georgeatparallel:georgeatparallel/webmcp-browser-tools
Draft

WIP: Add Parallel search and fetch tools for WebMCP sites#51
georgeatparallel wants to merge 10 commits into
parallel-web:mainfrom
georgeatparallel:georgeatparallel/webmcp-browser-tools

Conversation

@georgeatparallel

@georgeatparallel georgeatparallel commented Aug 26, 2026

Copy link
Copy Markdown
Contributor

This is a new way to distribute our free Search MCP: let websites offer it to the agents visiting them.

Today, an agent can browse a website without having Parallel available as a tool. Getting that access usually means the user has to configure our MCP server, or the team building the agent has to integrate us. WebMCP gives the website itself a way to offer tools to compatible agents while the page is open.

This PR adds @parallel-web/webmcp. A site owner adds an import or a script tag, and the package registers two tools with the browser: parallel_web_search and parallel_web_fetch. When an agent calls one, the package sends the request to our existing free Search MCP and returns excerpts with source URLs. The site doesn't need a backend or an API key, and the visitor doesn't need to configure our MCP server.

For example, a shopping site could let a visiting agent find independent product reviews or check compatibility across the web. The agent decides when to call the tools. Every site that adds the package becomes another place where agents can discover and use Parallel.

The scope is small: two read-only tools and a browser module with no runtime dependencies. It doesn't automatically collect page content or agent history. Requests omit browser credentials, and access from another origin requires the site to explicitly allow it. Browsers without WebMCP support safely do nothing. Sites still need to meet the browser's WebMCP requirements, which are covered in the README.

One browser limitation: Chrome 152 doesn't yet pass cancellation signals to tool callbacks, so cancelling a tool call there won't stop its network request. The package forwards the signal when the browser provides one.

Validation: 27 WebMCP tests, 311 total workspace and release tests, and all six CI checks pass. We also tested live search and fetch in Chrome, plus registration, request privacy, and cross-origin access.

Merging won't publish the package. An npm organization owner still needs to do the first release using the steps in PUBLISHING.md.

@georgeatparallel

Copy link
Copy Markdown
Contributor Author

Tracking this in DEV-259, assigned to me.

@georgeatparallel
georgeatparallel marked this pull request as ready for review September 4, 2026 02:29
@georgeatparallel
georgeatparallel requested a review from a team September 4, 2026 02:29
@georgeatparallel georgeatparallel changed the title Add browser-native Parallel Search tools for WebMCP websites Add Parallel search and fetch tools for WebMCP sites Sep 4, 2026
@georgeatparallel
georgeatparallel marked this pull request as draft September 4, 2026 02:51
@georgeatparallel georgeatparallel changed the title Add Parallel search and fetch tools for WebMCP sites WIP: Add Parallel search and fetch tools for WebMCP sites Sep 4, 2026
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant