mediar-ai/browser-cookies

Browser Cookies

Manage cookie persistence for isolated browser MCP agents. Isolated browsers load cookies from ~/.claude/browser-sessions.json at startup but do not save changes back automatically. This skill handles the export/import cycle.

Storage

• Cookie file: ~/.claude/browser-sessions.json
• Format: Playwright storageState JSON ({ "cookies": [...], "origins": [...] })
• Consumers: All isolated browser agents (isolated-browser, reddit-agent, twitter-agent, linkedin-agent) load from this file via storageState in their config

Workflow 1: Save cookies FROM an isolated browser

Use this after logging into a new service in the isolated browser. This exports the current session cookies and merges them back into the shared cookie file.

Step 1: Export cookies from the isolated browser

// Call via mcp__isolated-browser__browser_run_code
async (page) => {
  const cookies = await page.context().cookies();
  return JSON.stringify(cookies);
}

The result is a large JSON string. Save it to a temp file if it exceeds inline limits.

Step 2: Merge into browser-sessions.json

Parse the exported cookies and merge them into the existing file. When merging, cookies from the export take precedence (matched by domain + name + path).

import json

# Load exported cookies (from temp file or inline)
with open('/tmp/exported-cookies.json') as f:
    new_cookies = json.load(f)

# Load existing storage state
with open('~/.claude/browser-sessions.json') as f:
    state = json.load(f)

# Build lookup of existing cookies
existing = {}
for c in state.get('cookies', []):
    key = (c['domain'], c['name'], c.get('path', '/'))
    existing[key] = c

# Merge: new cookies overwrite existing, add any new ones
for c in new_cookies:
    key = (c['domain'], c['name'], c.get('path', '/'))
    existing[key] = c

state['cookies'] = list(existing.values())

with open('~/.claude/browser-sessions.json', 'w') as f:
    json.dump(state, f, indent=2)

Step 3: Restart the isolated browser

Close and reopen the isolated browser to load the updated cookies:

1. mcp__isolated-browser__browser_close
2. mcp__isolated-browser__browser_navigate to the target URL

Workflow 2: Seed cookies FROM the real Chrome browser

Use this to refresh all cookies from the user's real Chrome session (via playwright-extension). This replaces all cookies in the file.

Step 1: Export from playwright-extension

// Call via mcp__playwright-extension__browser_run_code
async (page) => {
  const cookies = await page.context().cookies();
  return JSON.stringify(cookies);
}

Step 2: Replace cookies in browser-sessions.json

import json

# Parse the exported cookies from the MCP result
# (handle the escaped JSON string from ### Result section)
new_cookies = ...  # parsed cookie array

with open('~/.claude/browser-sessions.json') as f:
    state = json.load(f)

state['cookies'] = new_cookies

with open('~/.claude/browser-sessions.json', 'w') as f:
    json.dump(state, f, indent=2)

Step 3: Restart isolated browsers to pick up new cookies

Workflow 3: Save cookies after EVERY isolated browser session

IMPORTANT: When you finish a browser task that involved any login, form submission, or authentication in the isolated browser, always run Workflow 1 before closing/leaving. This ensures the next isolated browser session has the updated cookies.

Notes

• Cookies are loaded once at context creation. Changes during a session are only in memory until explicitly saved.
• All isolated browser agents share the same browser-sessions.json. Saving from one agent updates cookies for all of them.
• Some httpOnly session-bound cookies may not transfer (e.g., GitHub user_session). These require re-login per session.
• Google session cookies (SID, HSID, SSID, APISID, SAPISID, etc.) DO transfer and persist well.