Markdown: images
Category: markdown
Renders  image references in text content blocks by fetching the URL. Distinct from base64 image content blocks.
Support by client
| Client | Status | Notes | Verified |
|---|---|---|---|
 Claude Web | unsupported | Click-to-reveal placeholder shown instead of inline image. Displayed: 'Here is a markdown image reference:' followed by a 'Show image' button (dark placeholder tile). The image did not render automatically — user must click. Call id: 1fafe8. | 2026-07-05T15:54:18.464Z |
| Claude Desktop | partial | Line 2: Markdown was parsed (no raw `![...]` syntax visible), but the image rendered as a "Show Image" placeholder button rather than the PNG displayed inline. | 2026-07-08T14:57:17.820Z |
| Claude Mobile | partial | Image ref parsed into a 'Load Image' placeholder card; tapping shows a navigate confirm then opens the PNG (nonce n=28382c in URL) in an in-app browser. Never renders inline in the reply. Automated via Mirroir. | 2026-07-09T14:30:03.848Z |
| Claude Voice | unknown | — | |
| Claude Code | partial | Tool was called and the model echoed the probe's call-id — but behavioral can't tell whether the client actually rendered the markdown variant (table/code/math/diagram/image/task/URL) vs. surfaced the source. Prefer vision-tier for a definitive answer. | 2026-07-08T19:16:52.012Z |
 ChatGPT Web | unsupported | Manual check (user-driven): the tool's markdown image reference was not rendered inline — no image appeared in the reply. Consistent with earlier pass where the model omitted the  line entirely (external image URL scrubbing). | 2026-07-05T15:37:38.337Z |
| ChatGPT Mobile | unknown | — | |
| ChatGPT Voice | unknown | — | |
| ChatGPT Desktop | unsupported | Manual check (user-driven): the tool's markdown image reference was not rendered inline in the ChatGPT Desktop reply — no image appeared. Same scrubbing behavior as chatgpt-web. | 2026-07-05T15:37:41.412Z |
| Codex CLI | unknown | Tool was called but the model did not echo the probe's call-id; can't infer rendering. Prefer vision-tier for a definitive answer. | 2026-07-02T17:50:17.475Z |
 Siri | unknown | — | |
| Apple AI (ChatGPT) | unknown | — | |
 Gemini Web | unknown | — | |
| Gemini Mobile | unknown | — | |
| Gemini Voice | unknown | — | |
| Gemini CLI | partial | Tool was called and the model echoed the probe's call-id — but behavioral can't tell whether the client actually rendered the markdown variant (table/code/math/diagram/image/task/URL) vs. surfaced the source. Prefer vision-tier for a definitive answer. | 2026-07-08T15:42:51.183Z |
 Antigravity | unknown | — | |
 Cursor | unknown | — | |
| Cursor CLI | partial | Tool was called and the model echoed the probe's call-id — but behavioral can't tell whether the client actually rendered the markdown variant (table/code/math/diagram/image/task/URL) vs. surfaced the source. Prefer vision-tier for a definitive answer. | 2026-07-04T14:45:06.872Z |
 VS Code (Copilot Chat) | unknown | — | |
 GitHub Copilot CLI | partial | Tool was called and the model echoed the probe's call-id — but behavioral can't tell whether the client actually rendered the markdown variant (table/code/math/diagram/image/task/URL) vs. surfaced the source. Prefer vision-tier for a definitive answer. | 2026-07-04T14:55:33.558Z |
 Windsurf | unknown | — | |
 Zed | unknown | — | |
 Perplexity Web | unknown | — | |
| Perplexity Mobile | unknown | — | |
| Perplexity Voice | unknown | — | |
| Comet Browser | unknown | — | |
 Grok Web | unknown | — | |
| Grok Mobile | unknown | — | |
| Grok Voice | unknown | — | |
| Grok CLI | partial | Tool was called and the model echoed the probe's call-id — but behavioral can't tell whether the client actually rendered the markdown variant (table/code/math/diagram/image/task/URL) vs. surfaced the source. Prefer vision-tier for a definitive answer. (tool_use event and debug log both missed; inferred toolCalled from call-id echo) | 2026-07-02T18:11:40.597Z |
 Warp | unknown | — | |
 Amp | unknown | — | |
 Goose CLI | partial | Tool was called and the model echoed the probe's call-id — but behavioral can't tell whether the client actually rendered the markdown variant (table/code/math/diagram/image/task/URL) vs. surfaced the source. Prefer vision-tier for a definitive answer. (stream event did not name the tool; inferred toolCalled from call-id echo) | 2026-07-04T14:53:47.570Z |
| Antigravity CLI | partial | Tool was called and the model echoed the probe's call-id — but behavioral can't tell whether the client actually rendered the markdown variant (table/code/math/diagram/image/task/URL) vs. surfaced the source. Prefer vision-tier for a definitive answer. | 2026-07-04T14:53:56.190Z |
 JetBrains AI Assistant | unknown | — | |
 Ollama | unknown | — | |
 LM Studio | unknown | — |