VSCode extension webview: 'Unsupported content type: server_tool_use / advisor_tool_result' when advisor tool is invoked

Status Closed — not planned
Reported on v2.1.138
Maintainer reply None cached
Activity 2 comments · opened May 19, 2026 · closed Jun 18, 2026

Summary

When the model invokes the advisor tool inside the VSCode extension, the webview renders two literal error strings instead of the tool call/result:

Unsupported content type: server_tool_use
Unsupported content type: advisor_tool_result

The advisor call itself succeeds — the model receives and acts on the response. Only the UI rendering is broken.

Environment

  • Claude Code CLI: 2.1.138
  • VSCode extension: anthropic.claude-code 2.1.143 (win32-x64)
  • OS: Windows 10 Pro 19045
  • VSCode: standard build (not Insiders)

Reproduction

  1. In any project, ask Claude to call the advisor tool (e.g. "call advisor()").
  2. Watch the conversation panel as the response streams.

Expected: the advisor invocation and its response render the same way other tool calls do (collapsible block, formatted result).

Actual: two plain-text "Unsupported content type" lines appear where the invocation and result should be.

Root cause

In webview/index.js the content-block switch handles text, image, document, tool_use, tool_result, and thinking, then falls through to:

return F4.default.createElement(
  "div",
  { className: UF.unknownContent },
  "Unsupported content type: ",
  F4.default.createElement("code", null, $.content.type)
)

The advisor tool emits server-side blocks (server_tool_use and advisor_tool_result, analogous to web_search_tool_use / web_search_tool_result) which are not in the switch, so they hit the fallback.

Suggested fix

Add cases for server_tool_use and advisor_tool_result (and likely the other server-side tool families) to the renderer — at minimum returning null or a quiet "Advisor consulted" pill so users don't see raw error strings.

Workaround

None in the UI — the errors are cosmetic and can be ignored; the advisor response still influences the model.

View original on GitHub ↗

This issue has 2 comments on GitHub. Read the full discussion on GitHub ↗