Connect Claude, Cursor, or Codex to the NEXUS AI MCP server and let agents deploy apps, read logs, back up databases, and roll back, with scoped OAuth access.
The server is remote and uses OAuth. Add https://mcp.nexusai.run/mcp to your client, sign in to NEXUS AI in the browser on first connect, and approve the scopes the agent may use. It is listed in the official MCP Registry as io.github.nexusrun/nexus-ai. Headless and CI clients can pass a pre-minted OAuth access token in the Authorization header instead.
Agents get tools for identity, AI Builder handoff, deployments (create, redeploy, logs, health, scaling, rollback, start, stop, auto-destroy), secrets, custom domains, external databases, deployment database backups and restores, persistent volumes, S3-compatible buckets, managed databases, and support tickets. Results come back as structured JSON the agent can act on.
Every tool requires one of 25 OAuth scopes. Grant deployments:read and deployments:logs and an agent can diagnose an outage without changing anything. Deleting anything needs a separate :delete scope, and restores and schema fixes need db:admin, so you decide how far an agent can go.
Every tool carries a hint: 26 are marked read-only and the 15 that change or delete data are marked destructive, so MCP clients can ask before running them. SQL that writes or changes schema runs only after a preview and an explicit confirmed flag, and schema fixes are proposed before they are applied. Deploys, restores, database queries, schema fixes, and managed database changes are recorded in the audit log.
An agent can generate an app in chat, push it to the AI Builder for a live preview with nexusai_builder_push, pull back your edits, deploy it with a database, then watch status and logs until it is healthy. Preview deployments can be given an auto-destroy time, and the agent can extend or cancel it later.
Most hosting platforms now ship an MCP server, but they differ in reach. NEXUS AI covers the full lifecycle, including backups, restores, storage, and secrets, and scopes each tool separately. The table below lists what each platform's server covers, from each vendor's documentation.
Add the server with one command. Claude Code opens the NEXUS AI sign-in on first connect.
claude mcp add --transport http nexus-ai https://mcp.nexusai.run/mcpFor headless use, add --header "Authorization: Bearer <token>" with a pre-minted OAuth access token.
Add the server, then run the login. The token is kept in the Codex credential store, not in the config file.
codex mcp add nexus-ai --url https://mcp.nexusai.run/mcp
codex mcp login nexus-aiThe add command writes an [mcp_servers.nexus-ai] block to ~/.codex/config.toml.
Add the server URL in Cursor Settings under MCP, or commit .cursor/mcp.json. Claude Desktop uses the same block in claude_desktop_config.json.
{
"mcpServers": {
"nexus-ai": {
"url": "https://mcp.nexusai.run/mcp"
}
}
}Claude Desktop reads ~/Library/Application Support/Claude/claude_desktop_config.json on macOS and %APPDATA%\Claude\claude_desktop_config.json on Windows.
One request, several tools: the agent reads health and logs, backs up the database before changing anything, redeploys, and verifies.
The checkout app is returning 500 errors. Check its health and the last 200 log lines, find the cause, back up its database, then redeploy with the fix and confirm it is healthy.This flow needs deployments:read, deployments:logs, deployments:create, and db:admin. Grant only the read scopes if you want a diagnosis without changes.
| Setting | Value |
|---|---|
| Server URL | https://mcp.nexusai.run/mcp |
| Transport | Streamable HTTP (remote server) |
| Authentication | OAuth with PKCE; sign in in the browser on first connect |
| MCP Registry name | io.github.nexusrun/nexus-ai |
| Tools | 74 in 11 categories |
| Permissions | 25 OAuth scopes, one or more per tool |
| Category | Tools | Examples |
|---|---|---|
| Identity and discovery | 4 | nexusai_whoami, nexusai_projects_list, nexusai_usage_stats |
| AI Builder handoff | 2 | nexusai_builder_push, nexusai_builder_pull |
| Deployments | 15 | nexusai_deploy_source, nexusai_deploy_logs, nexusai_deploy_scale, nexusai_deploy_rollback |
| Secrets | 4 | nexusai_secrets_create, nexusai_secrets_update |
| Custom domains | 4 | nexusai_domains_add, nexusai_domains_verify |
| External databases | 8 | nexusai_db_inspect_schema, nexusai_db_query_preview, nexusai_db_propose_fix |
| Database backups | 7 | nexusai_db_backup, nexusai_db_restore_to, nexusai_db_backup_schedule |
| Volumes | 5 | nexusai_volume_create, nexusai_volume_attach |
| Buckets | 10 | nexusai_bucket_create, nexusai_bucket_file_upload |
| Managed databases | 11 | nexusai_managed_db_create, nexusai_managed_db_snapshot_create |
| Support | 4 | nexusai_support_ticket_create, nexusai_support_ticket_reply |
| Platform | What its MCP server covers |
|---|---|
| NEXUS AI | 74 tools: deploys, logs, scaling, rollback, databases, backups and restores, storage, secrets, domains, and support, each gated by one of 25 OAuth scopes |
| Vercel | Deploy code, read logs, and manage projects |
| Render | Services, static sites, cron jobs, databases, logs, metrics, and Postgres queries |
| Railway | A hosted server with a railway-agent tool |
| Lovable | Create projects, chat with Lovable, read files, query the database, and deploy, with access to every project the account can reach |
| Replit | Eight tools to create, find, inspect, update, and publish Replit apps |
A deployment platform that exposes deploying and operating apps as Model Context Protocol tools, so AI agents in Claude, Cursor, Codex, and other clients can deploy, read logs, and manage databases from chat or the IDE.
https://mcp.nexusai.run/mcp. It is a remote Streamable HTTP server that uses OAuth, listed in the official MCP Registry as io.github.nexusrun/nexus-ai.
Run claude mcp add --transport http nexus-ai https://mcp.nexusai.run/mcp, then approve the NEXUS AI sign-in in the browser on first connect. Ask Claude to call nexusai_whoami to confirm the connection.
In Cursor, add https://mcp.nexusai.run/mcp under Settings, MCP, or in .cursor/mcp.json. In Codex, run codex mcp add nexus-ai --url https://mcp.nexusai.run/mcp and then codex mcp login nexus-ai. In Claude Desktop, add the same URL to claude_desktop_config.json and restart.
No. Each client signs in with OAuth in the browser on first connect, and you approve the scopes. Headless and CI clients that cannot open a browser can pass a pre-minted OAuth access token in the Authorization header.
Approve only the scopes it needs. Every tool requires one of 25 OAuth scopes: for example, deployments:read and deployments:logs allow diagnosis without changes, deletes need the matching :delete scope, and restores need db:admin.
Only with the scopes to do it. The 15 tools that change or delete data are marked destructive so MCP clients can ask before running them, SQL writes and schema changes need an explicit confirmed flag after a preview, and a restore can go into a separate database first.
Yes. nexusai_deploy_auto_destroy can set, extend, reduce, or remove a deployment's auto-destroy time without restarting the app.
Yes. nexusai_builder_push sends a generated file set to the project's AI Builder session and returns a preview URL where the user sees the app running. nexusai_builder_pull brings the files, including edits made in the builder, back into the conversation before deploying.
Those servers focus on deploying and inspecting apps on their own platforms. The NEXUS AI server also covers database backups and restores, storage, secrets, domains, and managed databases, and each tool is gated by its own OAuth scope.
NEXUS AI is an agentic AI app builder and full-stack deployment platform. Explore the AI App Builder, learn more on the About page, read the documentation, or contact the team through nexusai.run/contact.
Start for free · Read the documentation · About NEXUS AI · Contact