Overview
Inject API credentials into local agent requests.
Latchkey is a command-line tool that lets AI agents make authenticated HTTP requests on your behalf. You authenticate once, and then any agent on your machine can call Slack, Linear, GitHub, or any supported service using latchkey curl — no embedded tokens, no custom integrations.
Access Latchkey here: https://github.com/imbue-ai/latchkey
How it works
You authenticate (or an agent prompts you to): An agent that needs access to a service can invoke
latchkey auth browser slack, which opens a browser window for you to log in. You can also set credentials manually withlatchkey auth set.Latchkey stores credentials encrypted in ~/.latchkey on your machine.
Agents use
latchkey curlinstead ofcurl: When making API calls, the agent runslatchkey curlwith standard curl arguments.Latchkey injects authentication automatically: It detects the service from the URL and adds the necessary headers or URL modifications.
So instead of an agent needing to manage tokens directly:
curl -X POST 'https://slack.com/api/conversations.create' \
-H 'Authorization: Bearer xoxb-your-token' \
-H 'Content-Type: application/json' \
-d '{"name":"something-urgent"}'The agent simply runs:
latchkey curl -X POST 'https://slack.com/api/conversations.create' \
-H 'Content-Type: application/json' \
-d '{"name":"something-urgent"}'Latchkey detects the Slack domain and injects the authorization header automatically.
Why Latchkey?
AI agents need to interact with dozens of services, each with its own authentication flow and token format. Latchkey solves this, but the motivation goes beyond convenience.
Practical benefits:
Agents don't need custom code or dedicated connectors for each service
No credentials in prompts, config files, or environment variables
Authentication doesn't interrupt agent workflows — agents can prompt users to authenticate on demand via
latchkey auth browser, then continue working
Standards over custom integrations:
Latchkey encourages agents to work directly with existing public APIs via HTTP, leveraging well-established tools and standards (HTTP, REST, JSON, curl). This means compatibility even with services that haven't built anything specifically for agents. For services that do want to support the agentic use case, it's easier to maintain one set of API endpoints and documentation rather than a separate HTTP API and MCP server.
Software freedom:
Many MCP connectors are either unavailable for certain services, restricted to specific agents, or offered as paid features. Latchkey takes a different approach: it's open source and works with any agent that can run shell commands. Whether you're using local models, self-hosted APIs, or building your own AI assistant, Latchkey doesn't require you to depend on any particular vendor, backend, or OAuth app. You stay in control of your own credentials and data.
Access Latchkey here: https://github.com/imbue-ai/latchkey
Last updated