
Dejo
A browser your AI agent remembers: learns a website once, then replays it with zero model calls
MCP serverDeveloper tools2 viewslaunched Oct 23, 2026
Comments
0rohit kashyapMaker
What improvements would make Dejo easier to integrate with your AI workflow?
About Dejo
Dejo is an open-source MCP server designed for browser automation with Claude Code, Codex, and Cursor. It learns a website in a first exploratory run, saving all steps as a trail for efficient replay without additional model calls.
On subsequent runs, Dejo replays the learned trail using one tool call, validating each page and control along the way. If a website changes, it identifies which step broke and only repairs that step, saving the fix for future runs.
Dejo stores its memory locally on your machine and does not store passwords. It supports multiple signed-in accounts simultaneously and is free under the MIT license.
How it was built with Claude
I built Dejo with Claude Code as my pair programmer, and I test it with Claude Code as the agent. Dejo itself never calls a model: your agent (Claude Code, Codex, Cursor) does the thinking, and Dejo keeps the browser and the memory.
Gallery
Overview
Who is it for?
Developers and AI builders using Claude Code, Codex, or Cursor
Problem
Browser automation agents waste tokens and time by re-exploring websites for every task
Solution
Dejo learns a website once, records the browsing trail, then replays it with one tool call and zero model calls
What makes it unique
It detects and repairs only changed steps on website updates, and stores all data locally without passwords
The maker
Maker · @rohitkashyapmrt



