Skip to content

v0.1.0 — Initial release

Choose a tag to compare

@emretheus emretheus released this 10 May 18:56
· 2 commits to main since this release

claude-remind-mcp is a Model Context Protocol server that searches your local Claude Code conversation history. The running Claude agent can recall past solutions, drill into a specific message, summarise a session, or paste a ready-to-run claude --resume command — without you re-explaining the problem from scratch.

Install

claude mcp add claude-remind -- npx -y claude-remind-mcp

No model downloads, no daemons, no native modules. Local-only, ~20 KB tarball.

What's in 0.1.0

Added

  • BM25 search over local Claude Code conversation history (~/.claude/projects/).
  • Four MCP tools: remind_search, remind_message, remind_session, remind_resume.
  • Persistent index at ~/.claude-remind/index.json with mtime-based incremental rebuild.
  • Conservative solvedHint (likely / unlikely / unknown) derived from session-end signals.
  • format: "compact" | "detailed" | "full" snippet length control (400 / 1000 / 2000 chars).
  • Built-in secret redaction for common credential formats (OpenAI, Anthropic, GitHub, AWS, Stripe, Google, Slack, JWTs, private key blocks, common env assignments).
  • Symlink-escape guard during indexing.
  • Atomic index persistence (write-temp-then-rename) with 0600 file mode and 0700 directory mode.
  • Concurrent refresh guard.
  • Configurable via CLAUDE_CONFIG_DIR and CLAUDE_REMIND_DIR (absolute paths only).

Security

  • Per-line byte cap in JSONL parser to prevent memory exhaustion from oversized lines.
  • Per-file caps (5,000 docs, 50,000 messages) to bound index size.
  • Stripping of injection-style tags (<system>, <user>, <assistant>, ChatML, Llama [INST] / <<SYS>>) from indexed content before it returns to the calling agent.
  • Sanitized error messages (home directory and long file paths replaced).

Links

This is the first public release; expect minor breaking changes between 0.x versions. Issue reports and feedback welcome.