Home / Integrations

KeyLore supports common AI coding clients that need credentials during local development. The exact configuration differs by client, but the operating model stays the same: agents search metadata, choose the appropriate credential for the task, and work through the broker rather than reading raw tokens from a .env-loaded process.

Codex

Codex MCP setup

Connect KeyLore through Codex config and keep credentials outside the agent process.

Gemini CLI

Gemini MCP setup

Use KeyLore as the credential boundary instead of exporting tokens into the shell.

Claude CLI

Claude MCP setup

Register KeyLore once and let Claude request brokered access through MCP.

Universal pattern

The same four-step model across clients.

1

Start KeyLore locally

Run KeyLore on the local machine and open the UI to add credentials.

2

Define metadata

Tell the client what each credential is for so it can distinguish between services, environments, and intended use cases.

3

Configure the client

Point Codex, Gemini CLI, Claude CLI, or another MCP client at KeyLore using the client’s normal local config path.

4

Verify brokered access

Confirm the client can discover credentials and complete tasks without receiving raw token values directly.

How it works

Read the MCP guide

See what the model actually sees and how KeyLore expects agents to select credentials.

Security context

Read the MCP risk model

Understand why brokered credentials are safer than direct token exposure in agent tooling.