Skip to content

HANDOFF.md

Latest commit

 

History

History
156 lines (110 loc) · 4.01 KB

HANDOFF.md

File metadata and controls

156 lines (110 loc) · 4.01 KB

Midjourney Agent Handoff

This folder is a standalone local service for Midjourney Explore search and feed access.

It is intentionally decoupled from dnews runtime:

  • no dnews queue
  • no dnews backend API
  • no dnews database dependency

The service runs locally, opens a persistent Chrome profile, and uses the logged-in browser context to query Midjourney Explore APIs.

What It Can Do

  • report browser/login health
  • open the Midjourney Explore page in Chrome
  • search Explore images by keyword
  • fetch styles_top
  • fetch video_top

Proven Endpoints

These are the Midjourney endpoints verified during local testing:

  • GET /api/explore-vector-search?prompt=<keyword>&page=<n>&_ql=explore
  • GET /api/explore-srefs?page=<n>&_ql=explore&feed=styles_top
  • GET /api/explore?page=<n>&feed=video_top&_ql=explore

Current Local API

  • GET /health
  • GET /api/login/status
  • POST /api/browser/open
  • GET /api/explore/search?prompt=red&page=1
  • GET /api/explore/styles-top?page=1
  • GET /api/explore/video-top?page=1

Search Result Notes

  • Search result items do not include a detail page URL field directly.
  • A usable detail page can be derived as:
https://www.midjourney.com/jobs/<id>

Example:

https://www.midjourney.com/jobs/bd6d4e48-8a7c-4d18-b14a-0b62f1fe45f0

Why It Works

Midjourney blocks plain backend scraping with Cloudflare and auth checks.

The stable path is:

  1. launch a persistent Chrome profile
  2. keep the user logged in interactively
  3. open Midjourney Explore in that profile
  4. execute same-origin fetch calls from the page context

Earlier DOM-driven search automation was more fragile. The current implementation avoids depending on the search input DOM and calls the verified Explore endpoints directly from the logged-in browser page context.

File Map

  • package.json: project metadata and scripts
  • src/server.ts: local HTTP server and route handling
  • src/browser.ts: Midjourney browser session and API fetch logic
  • src/config.ts: runtime config parsing
  • src/types.ts: response and status types
  • src/utils.ts: helper functions
  • .env.example: environment template
  • README.md: quickstart

Setup

cd midjourney-agent
npm install
cp .env.example .env
npm run start

Recommended Runtime

Use a dedicated Chrome profile for the agent.

Recommended env:

MJ_USER_DATA_DIR="$HOME/.dnews-midjourney-profile/explore-debug"
MJ_HEADLESS=false

MJ_HEADLESS=false is recommended because:

  • first-time login and Cloudflare challenge are easier to complete
  • debugging is simpler
  • some flows are more reliable with a visible browser

First Run

  1. start the service
  2. call POST /api/browser/open
  3. complete Midjourney login / Cloudflare in the Chrome window if needed
  4. call search

Example:

curl -X POST http://127.0.0.1:18123/api/browser/open
curl "http://127.0.0.1:18123/api/explore/search?prompt=red&page=1"

Verified Local Result

search=red page=1 was verified successfully.

The service hit:

https://www.midjourney.com/api/explore-vector-search?prompt=red&page=1&_ql=explore

and returned 200 OK.

Deployment Notes For Air Bot

This should be deployed as a local or workstation-side bot, not as a pure remote backend crawler.

Reason:

  • Midjourney login is tied to a real browser session
  • Cloudflare challenges need browser state
  • credentials and cookies live in the Chrome profile

Deployment expectations:

  • macOS preferred
  • Google Chrome installed
  • interactive browser access available
  • persistent disk for the Chrome profile directory

Known Constraints

  • No task queue yet
  • No persistence layer yet for search results
  • No admin UI yet
  • loginState is heuristic, based on current page content/url
  • Midjourney may change DOM or endpoint contracts

Suggested Next Steps

  1. add normalized output format with detail_url
  2. add file/SQLite persistence
  3. add batch pagination fetch
  4. add simple auth token in front of the local API if needed
  5. add one-click startup shell script for operators