Home / Docs

KeyLore is designed to be understandable from first principles. Install it locally, add credentials, expose only useful metadata to AI clients, and let tools request brokered access when needed.

Getting started

Install and start locally

Get a local KeyLore instance running with the default npm install and HTTP start flow.

Product use

Use the local UI

Open the browser UI, add credentials, and define metadata that helps AI tools select the right secret.

MCP

Connect MCP clients

Understand how KeyLore fits into MCP-based workflows so agents can search metadata and request brokered access.

Security

Adopt a safer default than .env

Review why separating metadata from secret values is a better fit for AI-assisted coding than loading everything into process env.

Recommended first run

The shortest route to success.

  • Install the package and start the local HTTP service.
  • Open the UI at http://127.0.0.1:8787/ and add at least one credential.
  • Connect your AI coding tool or MCP client and verify it can discover metadata without receiving the raw token.
1

Install and start

npm install -g @simonsbs/keylore then keylore-http start.

2

Add one token

Store the secret, then write human context and LLM context so the agent can choose correctly.

3

Test and connect MCP

Use the built-in token test, then connect Codex, Gemini CLI, Claude CLI, or another MCP client.

Deeper reference

Full API and MCP docs on GitHub

Use the GitHub-hosted reference docs when you need endpoint-level and protocol-level detail.

Advanced operations

Production deployment notes

PostgreSQL, external secret stores, and advanced deployment concerns remain documented in the repo.