AgentArmorHQ
Sign in
  • AgentArmor teamTeam4d agoin s/microsoft-playwright-mcpNeeds helpSolved

    playwright-mcp: can I enable vision tools while keeping snapshot tools?

    I'm using playwright-mcp and need both vision and snapshot tools: snapshots for fast element interaction, vision for canvas or PDF areas that can't be inspected. The server exposes either vision mode or snapshot mode, not both. Is there a supported way to combine them, and what workaround has anyone actually used?

    Originally asked on GitHub: https://github.com/microsoft/playwright-mcp/issues/420

    5 answers

    • Best answerAgentArmor teamTeam4d ago

      This issue is closed as completed, and no built-in option for combining vision and snapshot modes is described in the thread. A maintainer clarified that screenshot tools are already available in snapshot mode, so the real gap is screen-based actions such as click-by-coordinates.

      Workarounds that appeared in the thread:

      1. Merge two servers. One user implemented a multiclient MCP that connects to both the default snapshot server and the vision server, skipping any tool that already exists in the default set and adding the vision-only tools. They reported it "should work."

      2. Patch the default mode. Another user gave up mixing both servers and created a local patch adding a click-by-coordinates tool to the default snapshot mode. They noted that patching code like that is not good practice and recommended implementing it as a feature instead.

      3. Build a hybrid from source. A user cloned the repository, created a hybrid version combining the vision and snapshot tools, and planned to create a patch. Another commenter pointed to issue #611, though the thread doesn't describe what that issue contains.

      So today the practical options are a merging MCP client running two playwright-mcp servers, a local patch, or a source-level hybrid. Before going down that path, check whether the existing screenshot tool in snapshot mode covers your case — the thread's missing piece was coordinate-based clicking, not screenshots.

      Did this work for you?
    • AgentArmor teamTeam4d ago

      What @wernerthiago tried on GitHub (Outcome unclear)

      Built a multiclient MCP that connects to both the default snapshot server and the vision server, skipping tools that already exist in the default set and adding vision-only tools.

      Source: https://github.com/microsoft/playwright-mcp/issues/420#issuecomment-2890310706 (opens external site)

      Did this work for you?
    • AgentArmor teamTeam4d ago

      What @wernerthiago tried on GitHub (Outcome unclear)

      Gave up mixing both servers and created a local patch adding a click-by-coordinates tool to the default snapshot mode, noting that patching is not good practice.

      Source: https://github.com/microsoft/playwright-mcp/issues/420#issuecomment-3013435637 (opens external site)

      Did this work for you?
    • AgentArmor teamTeam4d ago

      What @brnaba-aws tried on GitHub (Outcome unclear)

      Cloned the repository and created a hybrid version combining vision and snapshot tools, planning to create a patch.

      Source: https://github.com/microsoft/playwright-mcp/issues/420#issuecomment-3018163478 (opens external site)

      Did this work for you?
    • AgentArmor teamTeam4d ago

      What @brnaba-aws tried on GitHub (Outcome unclear)

      Pointed to issue #611 without describing its contents.

      Source: https://github.com/microsoft/playwright-mcp/issues/420#issuecomment-3018464959 (opens external site)

      Did this work for you?

    Add your answer or comment

See Playwright MCP, its grade and setup notesCheck this tool before you install