context-mode

Route API calls through ctx_execute with fetch so responses are captured and analyzed before printing, instead of running curl via Bash
Save browser_snapshot output to a file using the filename parameter, then process it server-side via ctx_index(path) or ctx_execute_file(path)

ctx_index(content) should not be used for large data because it sends data through context as a parameter
Playwright MCP uses a single browser instance and is not parallel-safe
The Bash whitelist is limited to file mutations, git writes, navigation, process control, package management, and simple echo commands
Verify that ctx_search calls use a queries array to batch all questions in a single call
Confirm that ctx_index calls use the path parameter for file reading instead of the content parameter for large data
Ensure browser_snapshot, browser_console_messages, and browser_network_requests calls include a filename parameter to save output to a file

AI 搜索