@oddeye/naver-blog-mcp

MCP server for Naver blog automation via Playwright. Use with Claude Desktop, Claude Code, Cursor, or any MCP-compatible agent to publish, schedule, and delete posts on Naver blog.

⚠️ Naver actively blocks automation. Session reuse works best; ID/PW login often triggers CAPTCHA. Use naver_login_interactive for first-time login.

Install via MCP client

Claude Desktop

Add to claude_desktop_config.json (macOS: ~/Library/Application Support/Claude/, Windows: %APPDATA%\Claude\):

macOS / Linux

{
  "mcpServers": {
    "naver-blog": {
      "command": "npx",
      "args": ["-y", "@oddeye/naver-blog-mcp"]
    }
  }
}

Windows (Node's spawn cannot resolve npx.cmd directly — wrap with cmd /c):

{
  "mcpServers": {
    "naver-blog": {
      "command": "cmd",
      "args": ["/c", "npx", "-y", "@oddeye/naver-blog-mcp"]
    }
  }
}

Claude Code

Add to .mcp.json in your project root, or run:

# macOS / Linux
claude mcp add naver-blog -- npx -y @oddeye/naver-blog-mcp

# Windows
claude mcp add naver-blog -- cmd /c npx -y @oddeye/naver-blog-mcp

Cursor

Add to ~/.cursor/mcp.json (use the same OS-specific command/args as Claude Desktop above).

First run downloads Playwright's Chromium (~150 MB). This may delay the first MCP startup by 30–90 s. If your client times out, install Chromium ahead of time:

npx playwright install chromium

Then restart your MCP client.

Tools

| Tool | Description | |---|---| | naver_login_with_credentials | Auto-login with ID/PW. ⚠️ Often blocked by CAPTCHA. | | naver_login_interactive | Open a browser window for the user to log in manually (handles CAPTCHA/2FA). Optional password arg stores credentials encrypted for auto session recovery. | | naver_login_status | Check whether a saved session exists for a given Naver ID. | | naver_register_account | Register an account in the local DB without logging in. | | naver_list_accounts | List registered accounts. | | naver_delete_account | Remove an account from the local DB. | | naver_create_post | Create a draft or scheduled post in the DB. | | naver_publish_post | Publish an existing DB post to Naver. | | naver_publish_now | Create + publish in one call. | | naver_publish_adhoc | Publish without writing to the DB (bypasses DB lookup). | | naver_list_posts | List posts (optional status filter). | | naver_delete_post | Remove a post from DB and (if published) from Naver. |

First-time setup flow

1. Tell the agent: "Register my Naver blog. ID: myid, blog URL ID: myblog_url. Use interactive login."
2. Agent calls naver_login_interactive. A Chromium window opens.
3. You log in manually (handles CAPTCHA / 2FA / new-device verification).
4. On successful login the session is saved to disk; account upserted.
5. Subsequent calls (naver_publish_now, etc.) reuse the session — no browser interaction needed.

Session resilience

If both ID and encrypted password are stored (via naver_login_with_credentials, or naver_login_interactive with the optional password arg, or naver_register_account with password), publish automatically recovers from session loss:

• No session file → auto-login with stored password before publishing.
• Session file exists but Naver invalidated it (SESSION_EXPIRED at publish time) → auto-login + retry publish once.

This means: give the credentials once, and the agent keeps publishing across long gaps without re-prompting the user. If you only saved a session (no password), session loss requires another interactive login.

Data location

Everything is stored under ~/.blog-automation/:

~/.blog-automation/
├── app.db                 # SQLite (accounts, posts)
├── sessions/<naverId>.json  # Playwright storage state
└── .encryption-key        # AES-256-GCM key (auto-generated on first run)

Override the base directory with BLOG_AUTOMATION_HOME.

Environment variables (all optional)

| Var | Default | Purpose | |---|---|---| | BLOG_AUTOMATION_HOME | ~/.blog-automation | Base data directory | | DATABASE_URL | $HOME/app.db | SQLite file path | | SESSION_DIR | $HOME/sessions | Playwright session storage | | ENCRYPTION_KEY | auto-generated to file | 64 hex chars (32 bytes) for password encryption | | HEADLESS | true | Set false to see the browser (publish/login) | | MCP_HTTP_TOKEN | unset | If set, the optional HTTP transport requires Authorization: Bearer … | | PORT | 3000 | HTTP transport port (if you run the bundled REST/HTTP server) | | SCHEDULER_CRON | * * * * * | Cron expression for scheduled-post polling | | SKIP_PLAYWRIGHT_INSTALL | unset | Set to 1 to skip Chromium auto-download during npm install (then run npx playwright install chromium manually) |

Security notes

• Passwords are encrypted with AES-256-GCM and stored locally only. They never leave your machine.
• If you pass a password through the agent's chat, it may be logged by the LLM provider. Prefer naver_login_interactive when possible.
• Naver detects automation aggressively from datacenter IPs. Run on your local machine or a Korea-region VPS for best stability.

Troubleshooting

Failed to reconnect to naver-blog: -32000 (Claude Code/Desktop)

• Windows: confirm the config uses "command": "cmd", "args": ["/c", "npx", ...] (not bare "npx"). Node's spawn cannot resolve .cmd shims without the wrapper.
• npx cache corruption: npm cache clean --force then restart the client.
• Startup timeout on first run: install Chromium ahead of time (npx playwright install chromium).

Unsupported URL Type "workspace:" during install

• Happens when running npm install @oddeye/naver-blog-mcp from a pnpm workspace cwd. Install in a plain folder, or use npx -y directly.

browserType.launch: Executable doesn't exist

• Chromium auto-install was skipped or failed. Run:

  npx playwright install chromium

Login keeps failing with CAPTCHA

• Use naver_login_interactive instead of naver_login_with_credentials. Naver blocks scripted ID/PW logins aggressively.

License

MIT