WebMCP

Try all 42 browser tools

Open the public native WebMCP demo. It registers 42 tools automatically using isolated sample data and simulated Network behavior. No account required. Directory scanners should label this route as a demo; the main app registers 11 tools before consent.

Your personal context, available to your agent

Keep facts, preferences, goals, and notes in an editable profile for AI browsers and agents. Share selected sections, export Markdown, and review suggested updates.

Browser tools

get_profile_template

Read the public starter template. No personal data.

get_profile_context

Read only profile sections approved for this tab.

search_profile_context

Find matching words in approved section titles and text. Returns complete matching sections with IDs and revisions.

get_task_context

Read selected approved sections for a task, or search their text. Taste context is optional and requires its own permission.

propose_profile_update

Propose a section edit or a new section. The user must approve it.

Save a collection for your next task

In your library, select sections, preview them, and choose Save as collection. Collections reuse current source text. Choose Use with AI to review the selection before sharing; a saved collection does not grant access.

Connect

1. Save some context in Personal context.
2. Select Use with AI, then Browser WebMCP, and choose sections.
3. Preview the selected text, continue, then select Allow in this tab.

For example, ask your agent to call search_profile_context with {"query":"goals","limit":5}. Only approved sections participate. If nothing matches, try words in your saved text. After a correction, call the tool again to read the current approved text in this tab.

For server access, follow the remote MCP setup guide. Browser WebMCP uses tools in the open tab. Export files creates a snapshot, not a connection.

Build with the API or MCP

Backend services and native apps can read approved context through the personal context JSON API. Compatible AI clients can use the hosted MCP server. Both use the same OAuth connection and permissions. API reads cover personal context search, profiles, task context, taste, catalog search, connection status, update status, and separately approved candidate comparisons.

Read the OpenAPI specification. The API also supports reviewed proposals and separately permitted save requests. A new integration needs the owner’s consent.

Privacy

Tools can be discovered without an account. Private reads require approval. No profile is published at a public file URL. Browser suggestions require review. Remote connections can save only with separate permission. See AI sharing.

Browser support

WebMCP is an experimental browser feature. The Google origin-trial token has an encoded expiry of 17 November 2026. Availability also depends on your browser and trial status. Markdown export and remote MCP remain available.