LanderLab: No MCP yet

LanderLab ships an official MCP server at landerlab.io/mcp, giving AI agents direct access to my landing page workflows. I also expose a REST API documented at landerlab.io/api, with an OpenAPI spec published at landerlab.io/openapi.json for tooling and client generation. Between the MCP server and the REST endpoints, I can automate page creation, template selection, and campaign asset management from external systems. Authentication details aren't published in the public docs I pull from, so confirm the method (OAuth or API key) before wiring production integrations. There's no Zapier or Make connector listed, no native ChatGPT or Claude integration, and webhook availability isn't documented; plan integrations around the MCP server and REST API directly.

Auth is one header: X-API-Key, with keys that start with ll_live_. No OAuth dance, no token refresh, no separate MCP credential to manage. You generate the key once and the same key works across REST and MCP. The server endpoint is https://api.landerlab.dev/api/v2/mcp, served over HTTP with Server-Sent Events so the agent keeps a live stream open while it runs a task.

The 30+ tools split into two groups, and that split is the part I like. Read-only tools (workspaces_list, landers_list, analytics_stats, leads_list, variants_list) let an agent survey the account, pull stats, and read leads without any risk of changing live pages. Mutation tools (lander_publish, editor_save, variants_set_weights, leads_update, integration_create) let it actually ship: save editor changes, publish a lander, reweight a split test, update a lead, wire an integration. The clean read/write boundary makes it easy to start an agent in a safe read-only posture and grant write access once you trust the workflow.

The client list is broad for a tool this size: Claude (Code, Desktop, and claude.ai), Cursor, Windsurf, Manus, ChatGPT, Gemini CLI, and any MCP-compatible client. Manus and Gemini CLI showing up by name is a nice signal that the server was tested beyond the usual Claude-plus-Cursor pair. Worth knowing up front: Grok, Perplexity, and Copilot chat clients cannot connect, and there is no LanderLab CLI, so the surface is agent clients only.

Workflows that pay off once it is wired: tell an agent to clone the winning variant of a split test and shift traffic weights toward it (variants_list reads the current split, variants_set_weights writes the new one), so you reweight without opening the dashboard. Or have Claude pull analytics_stats across your landers, find the underperformer, edit the headline through editor_save, and publish with lander_publish in one pass. Or run a lead sweep: leads_list to pull recent submissions, leads_update to tag or score them, integration_create to push them onward. The read-then-write pattern means you can keep an agent read-only while you watch it work, then let it ship once you trust it.