# MCP (AI clients)

> Manage your projects from Claude, Cursor and other AI clients over the Model Context Protocol.

Source: https://livops.dev/en/docs/mcp

Livops is an [MCP](https://modelcontextprotocol.io) server. Connect it to Claude Desktop, Claude Code, Cursor or any MCP client, and the model can run the cabinet from the chat: projects, stacks, deploys, variables, domains, databases, buckets, files, cron, team members, billing, hooks and notifications.

There are two ways in: through the browser (MCP OAuth) or with an [API token](https://livops.dev/en/docs/api-tokens). Either way the model works in one team, with your role in it. An app connected by someone with the **viewer** role only reads.

Creating, changing, stopping, deleting, reading a database password or a bucket key, redeeming a promo code, and opening a Payme or Click top-up link work when you connect through the browser. An `lvp_` API token can deploy and read. It cannot delete, reveal a password, or pay, so a leaked CI key cannot do those by calling a tool. The cabinet's own assistant is unchanged: it still does not delete.

## Address

```
https://livops.uz/api/mcp
```

Streamable HTTP.

## Connect through the browser

Add just the address to your client. On first connect Livops opens in the browser: you see the app's name, pick a team, see what the app will be able to do, and press **Allow**. With 2FA on, it asks for your code first.

The app gets a key that lasts an hour, works only at `/api/mcp`, and renews itself. Connected apps are listed under **Account → Connected AI apps**; press **Disconnect** and the app stops at once.

The app picks its own name and we don't verify it. If an app you didn't connect asks for access, press **Cancel**.

Clients whose docs describe this: Claude (as a custom connector — see below), Claude Code, Cursor, VS Code, Codex CLI, Gemini CLI and Zed. For Windsurf and Cline, use a token.

## With a token

Header `Authorization: Bearer lvp_...`. For CI, scripts and clients without OAuth.

## Claude Code

```bash
claude mcp add --transport http livops https://livops.uz/api/mcp
```

Then in Claude Code: `/mcp` → `livops` → **Authenticate**. With a token:

```bash
claude mcp add --transport http livops https://livops.uz/api/mcp \
  --header "Authorization: Bearer lvp_xxx"
```

## Cursor

`~/.cursor/mcp.json` (or `.cursor/mcp.json` in a project):

```json
{
  "mcpServers": {
    "livops": {
      "url": "https://livops.uz/api/mcp",
      "headers": { "Authorization": "Bearer lvp_xxx" }
    }
  }
}
```

## Claude

Livops is not in Claude's Connectors Directory. Add it yourself.

In claude.ai or Claude Desktop: **Customize → Connectors → Add custom connector**. Name it `livops` and paste:

```
https://livops.uz/api/mcp
```

Leave sign-in as the browser flow Claude detects. On first connect Livops opens so you can pick a team and press **Allow**. It then shows in your connectors list with a Custom label.

On a Team or Enterprise plan, an Owner adds it first under **Organization settings → Connectors** (Add → Custom → Web). Members then connect it themselves.

Claude Desktop can also start the [CLI](https://livops.dev/en/docs/cli) over stdio. In **Settings → Developer → Edit config** (`claude_desktop_config.json`):

```json
{
  "mcpServers": {
    "livops": {
      "command": "livops",
      "args": ["mcp"],
      "env": { "LIVOPS_TOKEN": "lvp_xxx" }
    }
  }
}
```

If you already ran `livops login`, the `env` block can go. Use the full path to `livops` (for example `~/go/bin/livops`) if Claude cannot find it.

## Tools

Every tool has a title. Read-only tools set `readOnlyHint`. Tools that change data set `destructiveHint`, so the client can ask you before they run.

### Projects and Git

| Tool | What it does |
| --- | --- |
| `list_teams` | teams you belong to |
| `create_team` | open another team |
| `switch_team` | move this connection to another team you already belong to |
| `list_github_repos` | GitHub repositories you can deploy (or GitLab, with `provider`) |
| `list_repo_branches` | branches of one repository |
| `list_projects` | the team's projects, status and address |
| `get_project` | one project: status, running service, domains, databases |
| `create_project` | create from a Git repository, a Docker image, or a preset |
| `update_project` | build command, start command, root, branch, health check, memory tier |
| `stop_project` | stop the running service until resume or the next deploy |
| `resume_project` | start a stopped service |
| `restart` | restart the running service |
| `delete_project` | delete the project |

### Stacks, environments, deploys

| Tool | What it does |
| --- | --- |
| `list_stacks` / `get_stack` | stacks and their environments |
| `create_environment` | staging (or another name) beside production |
| `delete_environment` | remove an environment that is not production |
| `rename_stack` / `delete_stack` | rename or delete a stack |
| `set_previews` | turn pull-request preview environments on or off |
| `promote_release` | promote a non-production release into production |
| `list_deploys` | deploys, newest first |
| `deploy` | redeploy from the current source |
| `cancel_deploy` | take a deploy out of the queue |
| `rollback` | put an earlier deploy back in service |
| `get_deploy_log` | build log, up to 5000 lines |
| `get_logs` | the running service's log, up to 5000 lines |
| `explain_failure` | why a failed deploy failed |
| `get_metrics` | CPU, memory and traffic for 1h, 24h or 7d |

### Variables, domains, data, storage

| Tool | What it does |
| --- | --- |
| `list_variables` | names, and which are secret — never values |
| `set_variable` / `set_variables` | create or change one, or several at once |
| `delete_variable` | delete one variable. Other secrets stay |
| `list_domains` / `add_domain` | custom domains and the DNS target |
| `verify_domain` | check DNS |
| `delete_domain` | detach a domain |
| `list_databases` | engine, service name, size — no password |
| `create_database` | Postgres, Redis, MySQL or MongoDB |
| `get_database_connection` | the connection URL, password included |
| `query_database` | one SQL statement or Redis command |
| `list_backups` | automatic snapshots. There is no "take one now" |
| `restore_backup` | restore into a new database. The original stays |
| `resize_database` / `delete_database` | a larger disk, or delete |
| `list_buckets` / `create_bucket` | S3 buckets beside a project |
| `get_bucket_credentials` | endpoint, access key and secret |
| `delete_bucket` | empty and delete a bucket |

### Billing

| Tool | What it does |
| --- | --- |
| `get_balance` | credit balance |
| `get_billing_summary` | plan and what it allows |
| `list_usage` / `list_payments` | ledger and payments |
| `redeem_promo` | put a promo code on the balance |
| `topup_link` | a Payme or Click checkout URL. No card number |
| `get_auto_topup` | whether the saved card tops up, and its last digits |
| `pause_auto_topup` / `resume_auto_topup` | stop or resume charging that card |

Backups are automatic. Stop is the customer's switch; an unpaid bill still suspends a project from billing, and no tool turns that off except paying. There is no tool that takes a database snapshot now.

## The rest of the cabinet

| Tool | What it does |
| --- | --- |
| `set_domain_redirect` | send a domain somewhere else, or clear that with an empty target |
| `plan_stack` | read a repository and say what a stack would contain. Creates nothing |
| `create_empty_stack` | a stack with production and no services |
| `add_stack_service` | add code or a database to a stack |
| `wire_services` / `rewire_stack` | connect two services, or write every address again |
| `merge_stacks` | move one stack's services into another |
| `list_bucket_files` / `upload_bucket_file` / `delete_bucket_file` | objects in a bucket. Upload returns a URL; you PUT the bytes there |
| `set_bucket_public` | public read on or off |
| `get_bucket_cors` / `set_bucket_cors` | CORS rules. `rules` is a JSON array |
| `get_bucket_lifecycle` / `set_bucket_lifecycle` | delete objects older than N days. 0 keeps them |
| `rotate_bucket_key` / `resize_bucket` / `get_bucket_usage` | a new key, a tier (`free`, `s`, `m`, `l`), daily usage |
| `list_database_users` / `get_database_usage` / `get_database_metrics` | who reads it, how full the disk is, charts for 1h, 24h or 7d |
| `list_database_extensions` / `set_database_extension` | Postgres extensions |
| `get_database_readonly` / `create_database_readonly` | the read-only URL, or a new password for it |
| `list_cron_runs` / `run_cron` | history, and run the saved command once |
| `move_project_address` | a new `*.livops.uz` name. The old one redirects for 30 days |
| `list_site_routes` / `set_site_routes` | paths on a static site answered by another service |
| `roll_push_secret` | a new push token |
| `switch_project_source` | point a folder or template project at a Git repository |
| `set_bot_mode` | polling or webhook |
| `set_wait_for_ci` | deploy after CI, or on every push |
| `get_protection` / `set_protection` | a password on the address. Empty clears it. The password is not read back |
| `get_site_analytics` / `set_site_analytics` | Google Analytics and Yandex Metrika ids |
| `check_seo` / `advise_seo` | the live site's SEO, and a written reading |
| `get_status_badge` / `create_status_badge` / `revoke_status_badge` | the README badge |
| `get_volume` / `set_volume` / `delete_volume` | the persistent disk. Sizes are 1024, 5120 or 20480 MiB |
| `list_team_members` / `invite_team_member` / `set_team_role` / `remove_team_member` | people in the team. Roles are member, viewer, billing |
| `rename_team` / `delete_team` | the label or the slug, or delete an empty team that is not the last one |
| `pay_charges` | a Payme or Click page for specific charges. `charges` is a JSON array of ids |
| `abandon_payment` / `get_receipt` | give up an unfinished payment, or read what a payment was for |
| `set_plan` | hobby, pro or team |
| `get_deploy_hook` / `create_deploy_hook` / `revoke_deploy_hook` | the secret URL that starts a deploy |
| `list_webhooks` / `create_webhook` / `delete_webhook` / `test_webhook` | outgoing deploy news |
| `get_log_drain` / `set_log_drain` / `delete_log_drain` | where logs are posted |
| `list_forms` / `list_form_submissions` / `mark_form_submission` / `delete_form_submission` | a static site's forms |
| `list_notifications` / `mark_notification_read` / `mark_notifications_read` | the bell |
| `get_telegram_link` / `unlink_telegram` | the notification chat |
| `list_presets` / `list_repo_folders` | templates for `create_project`, and folders for `root_dir` |

Shared environment variables stay closed to an `lvp_` token. These stay in the browser, with no tool: card numbers, the account password, two-factor, sessions, minting an API token, deleting the account, handing a project to somebody else, project members, downloading a raw backup, and anything an operator does.

## Safety

- Disconnect apps you no longer use under **Account → Connected AI apps**. With a token, make a separate one for the AI client and revoke it in **Settings → API tokens** when you stop using it.
- Logs, commit messages, query results and credentials are passed to the model as data. A log line that says "ignore previous instructions" is still just a log line — but your client decides what the model does next, so keep its confirmations on for tools that change things.
- A connection URL and a bucket secret are real secrets. Do not paste them into a shared chat.
- Calls are rate limited per account.
