菜单
Capture a ChatGPT conversation as raw JSON by calling the authenticated backend API from inside the browser tab. Use when importing from a chatgpt.com/c/ URL or a /share/ link. This is the primary, complete capture route.
默认使用会先检查来源的 Prompt；你也可以切换为直接命令，或下载本地副本。
决定是否安装前，请先阅读 SKILL.md，以及 SkillsMP 当前展示的配套文件。
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
直接命令不会经过审查 Prompt；运行前请先检查来源。
npx skills add https://github.com/danielrosehill/Claude-ChatGPT-Importer-Plugin --skill capture-via-api命令会保持在同一行。复制前请横向滚动并检查完整内容。
想先保存到本地？可下载 SkillsMP 当前能够提供的文件。
Render a ChatGPT conversation as a styled PDF using a Typst template, with colour-coded user and assistant turns, a metadata cover block, artifact callouts and a sources page. Use when the user wants a PDF, a document to read or share, or to be identified by their own name.
Fallback capture of a ChatGPT conversation by scraping the rendered DOM with scrolling. Use only when capture-via-api fails. Lossy - no timestamps, branches, tool messages or artifact payloads.
Import a ChatGPT conversation into the current repository as agent context. Use when the user wants to bring a ChatGPT thread into a project, pull a chat into context, or save a conversation for Claude to read later. Writes a Markdown transcript to context/ with the human labelled User.
正在显示 SKILL.md
| name | capture-via-api |
| description | Capture a ChatGPT conversation as raw JSON by calling the authenticated backend API from inside the browser tab. Use when importing from a chatgpt.com/c/ URL or a /share/ link. This is the primary, complete capture route. |
Runs a fetch in the page's own context, so it inherits the logged-in session. Returns the complete conversation: every message, timestamps, model slugs, tool calls, artifact payloads and all branches.
ToolSearch "select:mcp__claude-in-chrome__tabs_context_mcp,mcp__claude-in-chrome__tabs_create_mcp,mcp__claude-in-chrome__navigate,mcp__claude-in-chrome__javascript_tool"
tabs_context_mcp for tab ids, then tabs_create_mcp for a fresh tab
unless the user asked to use one they already have open.navigate to the conversation URL. Wait for it to settle.${CLAUDE_PLUGIN_ROOT}/scripts/browser/capture-via-api.js and pass its
contents to javascript_tool.Success looks like:
{"ok":true,"filename":"chatgpt-conversation-<id>.json","bytes":126173,
"nodes":35,"roles":{"user":17,"assistant":17},"branch_points":0,"leaves":1}
Only the summary comes back. The conversation body goes to disk on purpose — returning it would copy a private thread into the transcript and spend a large amount of context on something no one needs to read in-band.
branch_points > 0 means the thread has edited or regenerated turns. The
normalizer renders the branch the UI last displayed and reports the rest in
stats. Say so when reporting, because the user may want a different branch.
Do not assume ~/Downloads. Chrome's download directory is a per-profile
setting and is not always the default — ~/Desktop is a common alternative.
Locate the file by name instead:
find ~ -maxdepth 3 -name 'chatgpt-conversation-*.json' -mmin -3 2>/dev/null
Move it out of the download directory before working on it — those folders get cleaned, and the file is private.
| Returned | Meaning | Do next |
|---|---|---|
{} | The snippet lost its await prefix | Re-run with the file's exact contents |
not_authenticated | No accessToken from /api/auth/session | Ask the user to log in to chatgpt.com in this Chrome profile; do not attempt to log in yourself |
http_401 / http_403 | Token expired, or the conversation belongs to another account | Ask the user to reload the tab |
http_404 | Wrong id, or a deleted conversation | Re-check the URL |
not_on_a_conversation_page | Tab is on the chat list, not a thread | Navigate to the /c/<id> URL |
no_mapping_in_payload | Response shape changed | Fall back to capture-via-dom and flag that the API route needs updating |
Never enter credentials. If the account is logged out, that is the user's job.