Inquiry About Reusing MCP Instance for Sequential Operations in Playwright-MCP

[v-chenyiqiang]

Description:

I am working with the Playwright-MCP (SEE) service, utilizing a client codebase built on fastmcp. Here's a breakdown of my scenario:

1. Service Initialization: I start the Playwright-MCP (SEE) service and prepare to issue commands to a large model.

2. Initial Query: I send a request to the model: "Search Google for the local weather and retrieve the weather forecast based on the search results."

3. Model Callback: The large model, having already accessed my tool capabilities via MCP, proactively calls the Playwright-MCP API. The Playwright-MCP performs the following steps:

   • Opens a browser window.
   • Navigates to https://www.google.com.
   • Returns the current page results in a structured tree format.

4. Response from Playwright-MCP: The results are sent back to the large model.

5. Model's Next Input: The model then makes another input: "Get today's weather"

   ChatCompletionMessage(content=None, refusal=None, role='assistant', annotations=None, audio=None, function_call=FunctionCall(arguments='{"element":"textbox \\"河南启动对京东外卖的调查\\"","ref":"e35","text":"当地天气","submit":true}', name='browser_type'), tool_calls=None)
   

6. Execution Feedback from Playwright-MCP: The response from Playwright-MCP is:

   {'success': False, 'error': 'Error calling server tool: Error: No current snapshot available. Capture a snapshot of navigate to a new location first.'}
   

Clarification Needed:

I would like to confirm if the MCP server supports the ability to reuse the same instance for sequential operations. Specifically, in my scenario, can the model invoke the MCP API in a second call to perform actions directly on the already opened page from the first call? Thank you!