Krimto

What krimto-labs/krimto MCP server does

The krimto-labs/krimto MCP server gives AI coding agents a shared memory layer whose source of truth is a Git repository. Each stored fact is a Markdown file with metadata such as scope, title, author, timestamps, tags, and source. Developers can inspect, edit, commit, and review these files directly instead of relying on an opaque hosted memory database.

Memory can be personal to a user, shared with a team, or available across an organization. When more than one scope is relevant, the more specific scope takes precedence during recall. This supports solo use while providing a defined access model for team knowledge.

How it works

The system separates storage, retrieval, and access control. Markdown files in the Krimto data directory provide the durable record, and Git supplies synchronization and an audit trail. A hybrid SQLite and sqlite-vec index supports keyword and vector retrieval over that content. The access layer determines which user, team, or organization scopes can be read or written.

Agents communicate with the MCP server over stdio. The same server can be connected to supported coding environments, allowing a fact saved in one editor to be recalled in another. Git synchronization can also carry the Markdown memory to another machine. Team deployments may use one shared server or separate Krimto instances synchronized through a shared Git remote.

Setup and configuration

The npm package can initialize a local setup with:

npx @krimto-labs/krimto init

The setup wizard detects supported editors and configures them. Cursor and Claude Code are listed as automatic connections; Codex and Gemini CLI require a manual configuration snippet. A generic stdio configuration uses npx with the package name and the -y flag.

For team mode, run npx @krimto-labs/krimto team init. The wizard collects an administrator email, organization and team details, an optional shared Git remote, and teammate invitations. A self-hosted HTTP server and browser dashboard can be started with npx @krimto-labs/krimto serve, or the published Docker image can be run with the data directory mounted at /data.

The krimto-labs/krimto MCP server does not require an account for the solo setup. Data is stored under ~/.krimto by default. The project requires Node 20 or newer when run with Node.

Tools and capabilities

• Save durable facts from an agent conversation.
• Recall facts in later chats, editors, or machines.
• Store memory as reviewable Markdown files in Git.
• Retrieve information through keyword and vector indexing.
• Apply user, team, and organization scopes during access and recall.
• Inspect notes with the krimto notes command or the browser UI with krimto ui.
• Connect coding agents through the Model Context Protocol over stdio.

Limitations and notes

The provided material describes the MCP connection and command-line setup but does not enumerate individual MCP tool names. Codex and Gemini CLI need a manual connection step, while Cursor and Claude Code are configured automatically by initialization. The project roadmap lists OAuth sign-in and pull-request approval as future work, so those capabilities should not be assumed in the current version. A hosted Krimto Cloud service is also described as planned rather than currently available. Team mode requires additional setup and may use a shared server or a shared Git remote.