Skip to content

feat: Add headless browser MCP tools for AI agent web automation#22

Open
root3315 wants to merge 1 commit intogoogle:mainfrom
root3315:feat/browser-mcp-tools
Open

feat: Add headless browser MCP tools for AI agent web automation#22
root3315 wants to merge 1 commit intogoogle:mainfrom
root3315:feat/browser-mcp-tools

Conversation

@root3315
Copy link
Copy Markdown

@root3315 root3315 commented Apr 4, 2026

Implements #14

Adds 9 browser tools (navigate, click, type, screenshot, evaluate, getText, fillForm, getContent, close) backed by Playwright running on the host. Browser starts lazily on first tool call, maintains stateful session across calls.

Architecture

Node.js Playwright server runs on host, Rust wrapper manages process lifecycle via subprocess with HTTP API. All calls logged to session.db telemetry.

Key files

  • crates/capsem-core/src/mcp/browser_tools.rs (new, 1244 lines)
  • crates/capsem-core/src/mcp/playwright_server.js (new, 160 lines)
  • skills/dev-browser/SKILL.md (new documentation)
  • crates/capsem-core/src/mcp/gateway.rs (updated for browser routing)

Requirements

Node.js + Playwright on host:

npx playwright install

Testing

cargo test -p capsem-core mcp::browser_tools

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