Chat Recall

What chat-recall does

chat-recall brings AI coding-session history from multiple tools into one index. It can search conversations by their contents rather than only by filenames, expose recent work, and make earlier decisions, project facts, and conventions available across agents and teammates. Indexed sources can include sessions as well as skills, MCPs, commands, subagents, and hooks.

The server groups its capabilities into four areas:

• Shared, time-aware knowledge that can distinguish superseded decisions.
• Security findings for credentials found in past sessions, including checks of whether a credential still works.
• Ranked code-improvement recommendations that can be applied or dismissed with repository code intelligence.
• A shared task board with repository-specific rules.

Twenty-eight additional tools support searching, replaying, and comparing indexed sessions. In total, the server exposes 68 MCP tools.

How it works

The CLI reads transcripts already stored on the local machine and syncs the selected data to either the hosted service at chatrecall.dev or a self-hosted server. Search uses PostgreSQL full-text search. AI-generated summaries are optional and are not required for search.

Before an initial sync, chat-recall init displays the number of sessions, source tools, and project names that would be uploaded, then asks for confirmation. Secrets are masked on the client, and project paths are sent as hashes. Local rules can exclude projects, tools, or transcript sources; sync-only rules can instead define an explicit allowlist. Existing remote sessions remain until separately deleted, while retention rules can remove older server-side data on a schedule.

Setup and configuration

Run the initializer with:

npx chat-recall init

It indexes local transcripts, detects supported tools, and adds the MCP registration to configurations for Claude Code, Codex, OpenCode, and Cursor when those tools are present. It does not overwrite manually curated entries. chat-recall doctor reports registration status, and --skip-sync prevents the initial upload. The Claude Code plugin can install the skills and MCP server together.

Other MCP clients can use the stdio command directly:

{
  "mcpServers": {
    "chat-recall": {
      "command": "npx",
      "args": ["-y", "-p", "chat-recall", "chat-recall-mcp"]
    }
  }
}

The package name starts the CLI, so the MCP process must use the chat-recall-mcp binary. chat-recall search searches indexed history, and chat-recall recent lists recent work. Hosted use begins with a seven-day trial; self-hosting is free for one person and uses the same CLI with a different server URL.

Tools and capabilities

• Search sessions across supported AI coding tools.
• Replay and compare prior sessions.
• Share project facts, conventions, and decisions between agents.
• Detect credentials in historical conversations and check whether they remain active.
• Rank, apply, or dismiss repository code recommendations.
• Track findings as tasks with repository-specific rules.
• Exclude projects, tools, or transcript sources from synchronization.
• Delete individual sessions and configure server retention.

Limitations and notes

Local exclusions apply from the next sync; data already uploaded remains until deleted. Path-based rules cannot match sessions stored under hashed paths, so tool-level exclusions are available for those sources. Retention deletion can only be reversed by re-syncing the original transcript while it still exists locally.

The hosted service requires a paid subscription after its seven-day trial. Self-hosting keeps the service on the user’s own machine, while collaboration features such as shared history and assigning work require a license for additional members. No API key is required to run the CLI or use search.