Every AI coding session starts fresh. Switch tools, start a new thread, or come back after the weekend and the useful context from last week is gone. Recall is a new open-source tool that fixes this by indexing all your AI chat history locally and making it searchable with a single command.

What Happened

Developer Pratik Gajjar published Recall v0.1.1 on June 1, 2026. It indexes conversation files from four AI coding environments: Cursor, Claude Code, OpenAI Codex CLI, and pi. Everything goes into a local SQLite database using FTS5 full-text search with BM25 ranking. No data leaves your machine. The tool ships as a single pure Go binary around 10 MB with no runtime dependencies, making installation straightforward on macOS and Linux.

Why It Matters

AI assistants are stateless by design. Months of useful problem-solving is locked inside individual sessions and tools with no way to search across them. If you switched from Claude Code to Cursor last month, any context you built up in earlier sessions is effectively invisible in your current workspace.

The MCP server mode turns Recall into something more than a search tool. Run recall mcp and your entire conversation history becomes available as a Model Context Protocol server. Agents in Claude Code, Cursor, or Codex can pull relevant past sessions automatically as live context during new tasks, turning accumulated history from dead weight into an active resource.

Key Details

  • Commands: recall index builds the index. recall <query> searches. recall last shows the most recent session.
  • Local-first: Read-only indexing. Source files are never modified or uploaded anywhere.
  • MCP server mode: recall mcp exposes your history as an MCP server for agent integration.
  • Single binary: Pure Go executable, no CGO, no runtime dependencies.
  • Extensible: An adapter interface makes adding new tool support straightforward.
  • License: MIT.

If you already use Cursor's agent mode, wiring Recall in as an MCP server gives the agent access to your full session history before it starts each new task.

What to Do Next

Recall is available now at github.com/pratikgajjar/recall. Install and configuration instructions are in the README. The tool is at v0.1.1 so expect occasional rough edges, but the core search and MCP server modes are working. Watch the repo for updates as support for additional platforms gets added.