Install the Somewhere MCP server.
Database, auth, payments, email, file storage, custom domains, and deploys for any MCP-capable agent. Free to start, no credit card.
MCP is for connected clients. If your agent has a shell (Claude Code, Codex), it can use the CLI instead — no MCP connection and no account for a first, temporary deploy: see /start.txt.
Connector URL: https://mcp.somewhere.tech/mcp
Codex
OpenAI Codex CLI
Local bridge — recommended, auth rides your CLI login
npm i -g @somewhere-tech/cli
somewhere login
codex mcp add somewhere -- somewhere mcp
Remote (streamable HTTP) — if your Codex setup manages auth itself
codex mcp add somewhere --url https://mcp.somewhere.tech/mcp
Or let the CLI write the config for you
somewhere mcp install codex
Claude Code
Anthropic Claude Code CLI
Fastest — login configures Claude Code automatically
npm i -g @somewhere-tech/cli
somewhere login
somewhere login writes the MCP entry for you. No extra step.
Or let the CLI write the config for you
somewhere mcp install claude-code
Cursor
Cursor editor
One command
npm i -g @somewhere-tech/cli
somewhere login
somewhere mcp install cursor
Or manually — add to ~/.cursor/mcp.json
{ "mcpServers": { "somewhere": { "command": "somewhere", "args": ["mcp"] } } }
Claude.ai (web)
No terminal needed — the simplest install.
Connectors
Open Claude.ai → Customize → Connectors
Click + → Add custom connector
Paste https://mcp.somewhere.tech/mcp, then Connect
Any other MCP host
Continue, Windsurf, your own agent — anything that speaks MCP.
Remote (streamable HTTP)
URL: https://mcp.somewhere.tech/mcp
Auth: OAuth, or header Authorization: Bearer <your developer key>
Local stdio
npm i -g @somewhere-tech/cli
somewhere login
Run as the server command: somewhere mcp
How do I know it worked?
In a new agent session, the tool list should include names like:
mcp__somewhere__project_create
mcp__somewhere__db_query
mcp__somewhere__project_deploy
| Codex | codex mcp list |
| Claude Code | /mcp (inside a session) |
| CLI login | somewhere auth status |
| Everything at once | somewhere mcp doctor |
Remote vs. local bridge
Remote (streamable HTTP)
Your host connects straight to https://mcp.somewhere.tech/mcp and handles auth itself — OAuth on Claude.ai, or a bearer key on hosts that support custom headers.
Local bridge (stdio)
Your host runs `somewhere mcp`, which signs every request with your CLI login from ~/.somewhere/config.json. If `somewhere whoami` works, the bridge works — no token pasted into host config.
FAQ
What is MCP?
Model Context Protocol — Anthropic's open standard for connecting AI agents to external tools and data. The platform offers an MCP server, which means any MCP-compatible agent can use it to provision databases, auth, payments, email, file storage, and more — without writing boilerplate.
Which agents work with Somewhere?
Shell agents use the CLI; connected clients use MCP. Officially tested over MCP: Claude.ai (added as a custom connector with the URL above), Cursor, and Continue; Claude Code via the CLI. If your agent supports MCP, point it at https://mcp.somewhere.tech.
I installed it but my agent doesn't see the tools — is it broken?
Almost certainly not. Most MCP hosts load tools only when a session starts, so a server that shows as configured can still be invisible to the session you already had open. Open a new session (or restart the host) after any config change. Run somewhere mcp doctor to check login, connectivity, and host config in one shot.
Do I need a Somewhere account?
For MCP, yes — free to sign up, no credit card required. The full core platform — database, auth, email, custom domains — is on the Free tier; paid tiers just raise the limits and add scale and support. For a first deploy with the CLI, no: an agent with a shell can ship a temporary three-hour app while logged out and hand you a claim link (see /start.txt).
Can I use this with my existing Claude subscription?
Yes — that's the point. Your Claude Pro or Max subscription already covers the intelligence. Somewhere covers the infrastructure. You're not paying for Claude twice.
Can the agent write to my database and files?
Yes, but scoped to your project. The agent operates under your developer key — same scope as you. You can also set row-level scoping so end-user data is auto-isolated.
Can I see what the agent did?
Your dashboard shows activity and deployment history. You can request a rollback to a retained deploy, subject to database schema safety checks. In-place database restore is unavailable; export or download a copy before database changes.
What if I want to leave?
Full export. All your code, all your database, all your files. Download and run anywhere. No lock-in — we keep you because we are better, not because you are trapped.
Is there a surprise charge on AI?
No surprises. In-app AI, images, and TTS bill pay-as-you-go from a prepaid balance — you see each rate up front and can't be charged beyond what you've funded. Your connected agent runs on the subscription you already have.