Every write action (creating a project, resolving an issue, and so on)
requires the agent to pass an explicit confirmation flag, which a
well-behaved agent will only do after asking you. Read actions (listing
orgs, projects, issues) don’t require confirmation.
Connect Claude Code
Connect Replit
In Replit Agent: Integrations → MCP Servers → + Add MCP server, then paste:Other MCP clients
Any client that supports a remote MCP server over HTTP with OAuth 2.1 (dynamic client registration + PKCE) can connect the same way — point it athttps://mcp.lerix.dev/mcp and complete the sign-in when prompted. There’s
no API key to generate or paste in.
What it can do
Read (no confirmation needed):
Create / connect (requires confirmation):
Write (requires confirmation):
Security notes
- All access is scoped by your existing organization/project permissions — the agent can’t see or change anything your account couldn’t already.
- Nothing the agent connects to can read your password. Sign-in happens on
app.lerix.devitself, the same login you already use. - To disconnect a client, remove the server from it directly (e.g.
claude mcp remove lerix). If you believe a specific token was compromised, contact support@lerix.dev. configure_android_pushandconfigure_ios_pushask you to paste a real push-provider credential (a Firebase service-account key, or an APNs.p8key) into the chat. That text passes through your AI tool the same way any other message does — only paste it into a client and session you trust.