AI Agents & MCP
Ask your app data anything.
Connect your coding agent to Lotse over MCP. It reads the same funnels, feedback and campaigns you see in the console.
Works withClaude CodeCursorVS CodeCodex
What you can ask
Your project, in plain language.
“Which campaign brought the most buyers last month?”
“Where in onboarding do new users give up?”
“Summarise this week’s feedback by topic.”
“Did last week’s push improve day-7 retention?”
“Save everyone who abandoned the cart as a segment.”
“Create a link for the autumn newsletter.”
Safe by design
You decide what an agent may do.
Every token belongs to exactly one project and expires. Pick the lowest level that does the job.
- One project per token — no way to reach other data
- Expires after 30, 90 or 365 days
- Cannot send campaigns, cannot delete anything
Honest
Lotse runs no AI of its own.
Your agent, your AI provider, your choice. Lotse answers with your project’s data — the same numbers and time ranges as the console.
For developers
One command to connect.
claude mcp add --transport http --scope local lotse https://api.lotse.dev/mcp --header 'Authorization: Bearer ${LOTSE_MCP_TOKEN}'
- The token stays off the command line
- Typed without echo or read from the macOS keychain when the client starts.
- Clients
- Claude Code, Cursor, VS Code and Codex, with snippets in the console.
- Endpoint
- Streamable HTTP at api.lotse.dev/mcp.
Questions about AI agents
Does Lotse send my data to an AI provider?
No. Lotse runs no AI. Your agent calls Lotse; what it does with the answer is up to your client and provider.
Can an agent break something?
It cannot send a campaign and cannot delete anything. Actions that reach your users need a publish token.
Which clients work?
Any MCP client over HTTP. Accepted with Claude Code and Codex; snippets for Cursor and VS Code are in the console.
AI Agents & MCP
Give your agent the full picture.
Create a read-only token in the console and connect in two minutes.