Claude Code Context Capturer

Claude Code Context Capturer turns any web page into clean Markdown and writes it directly to your project's CLAUDE.md — closing the "I researched something for an hour, now my AI agent has none of that context" loop. Why use it? • URLs alone don't always work — many LLMs can't fetch external content reliably • Copy-pasting brings noise (ads, menus, footers) • Manual paste-into-CLAUDE.md doesn't survive busy days • Different projects need different context files • Multiple AI agents (Claude Code + Cursor + Windsurf) each want their own context file What it does: • Direct write to CLAUDE.md — pick the file once with the File System Access API; subsequent captures append to it automatically • Multi-project routing — link multiple CLAUDE.md files and route by URL glob (e.g. github.com/anthropic/* → one file, zenn.dev/* → another) • Multi-agent fan-out (v0.9.0+) — one route can write to multiple context files at once. Same capture lands in CLAUDE.md (Claude Code), .cursorrules (Cursor), and .windsurfrules (Windsurf) without setting up duplicate routes per agent • claude.ai conversation capture — capture your brainstorm conversations with Claude (thinking blocks, tool_use, branches preserved) straight into your project context • Artifacts-only capture & range selection (v0.5.0+) — for long claude.ai planning chats, capture just the code/documents Claude wrote or only the last N messages, toggleable from the popup • Re-capture updates in place (v0.5.0+) — re-capturing the same claude.ai conversation overwrites the existing capture instead of creating duplicate snapshots • YouTube transcript capture (v0.8.0+) — for tutorials, talks, and conference recordings: pulls the transcript (manual or auto-generated) plus title / channel / chapters / duration. Chapters become sub-headings with [mm:ss] timestamps • Preview before write (v0.6.0+) — review the captured Markdown in a window, trim cruft like nav menus or sidebar leftovers, then confirm. Optional, toggleable from a "skip next time" checkbox • MCP server for on-demand context (v0.5.0+) — save captures to a directory and the bundled MCP server hands them to Claude Code only when it asks for them, keeping CLAUDE.md small. Tools: get_context / list_contexts / search_contexts / stats_contexts, with tag/date filtering. v0.7.0 added the CCC_MCP_TOOLS env var to trim the tool set and save ~666 tokens/turn vs the full set • Site-specific parsers for GitHub, Stack Overflow, Zenn, Qiita, MDN • Generic parser (Mozilla Readability) for everything else • Selection mode: capture just what you've highlighted • Buffer mode: stack captures in extension storage for later export • YAML frontmatter with URL, title, captured_at, tags • One-click from toolbar / keyboard shortcut / right-click menu • English and Japanese UI Keyboard shortcuts: • Ctrl+Shift+L (Cmd+Shift+L on macOS) — capture page • Ctrl+Shift+K (Cmd+Shift+K on macOS) — capture selection Privacy first: • 100% local processing — nothing leaves your device • No telemetry, no analytics, no external servers • File handles stored locally in IndexedDB • Open source under the MIT license Source code: https://github.com/OceansCreative/claude-code-context-capturer