Not translated yet — showing English.
Backoffice MCP Integration
MCP lets an AI agent work in your project the way you do in the Backoffice. Connect an agent such as Claude or Cursor, and it can read your data, evolve your schema, manage players and the catalog, and deploy functions, by chatting in plain language. Everything is on AI Access under Engineering in the sidebar.
Model Context Protocol is an open standard for giving AI agents tools. PlayServ's MCP server offers the agent the same operations as the Backoffice, with the same permissions as the person who signs in.
What the agent can do
| Area | For example |
|---|---|
| From brief to backend | Stand up a whole project from a single brief: schema, catalog and functions. |
| Schema & migrations | Add fields, create entities, plan and apply safe migrations. |
| Data | Query, create and bulk-import records; manage relations and saved views. Records of a player-owned entity can't be created by the agent: they need a signed-in player. |
| Players | Look players up, manage sessions, suspend or ban. |
| Catalog & storefronts | Manage items and integrations, publish storefronts. |
| Functions | Inspect and deploy functions, manage secrets, read metrics and logs. |
| Analytics | Pull dashboards, baselines and events. |
| Projects & keys | Create projects, issue and rotate API keys. |
Built-in recipes walk the agent through multi-step jobs, such as setting up a project from scratch. The agent can also reach a few things the Backoffice doesn't show, like raw fields and pushing a schema from code.
Permission tiers
Every tool is Read, Write or Admin, so you can let reads run on their own and confirm the rest.
| Tier | What it does | Suggested approval | Endpoint |
|---|---|---|---|
| Read | Lists, gets, queries, dashboards, metrics, logs. No changes. | Auto-approve | /mcp/read |
| Write | Creates, updates, publishes, enables. Reversible changes. | Ask each time | /mcp/write |
| Admin | Deletes, key and session revocation, player bans, disconnects. Destructive. | Deny by default; approve each call | /mcp/admin |
Your MCP client groups the tools by tier on its own when you connect to /mcp. To handle the tiers separately, add each endpoint as its own server and set approval per server in the client.
Connect your client
Add the server to your client
Open AI Access → Connect MCP and copy the configuration into your client: Claude Desktop, Claude Code, Cursor or any other MCP client.
{
"mcpServers": {
"playserv": { "type": "http", "url": "https://platform.playserv.com/mcp" }
}
}In claude.ai, add it as a custom connector instead: under Settings → Connectors (or Organization settings → Connectors on a Team or Enterprise plan), choose Add custom connector, name it PlayServ and paste the URL.
Sign in
The agent's first PlayServ tool call opens a sign-in page. Sign in with the Google or GitHub account that has access to your PlayServ studio. The agent then acts with that account's permissions.
Check the connection
Ask the agent "What PlayServ tools do you have?". It should list them. To inspect the server without an agent, point the MCP Inspector at the same URL.
Never give the agent an API key
The agent signs in as you; it doesn't need your project's API keys, and a key pasted into a chat may be stored with the conversation. Keep keys in your game, servers and CI.
Pick a project
Tell the agent which project to work in by its slug, which is in the project's settings on Projects. The agent finds the project and uses it for the rest of the conversation.
Change the schema safely
Ask for changes in plain language; the agent describes what it is about to do before doing it. Review its plan before you confirm.
Small changes the agent can make directly, like a new object or an optional field, take effect at once. A change that would break existing records, such as removing a field or making one required on an entity that has records, is refused as a direct change: the agent plans it as a migration, shows you its impact, and applies it when you agree. Access settings changed by the agent apply at once, without the migration the Backoffice stages them in.
Generate a schema from a game design document
Paste a game design document, or any description of your game's data, and ask the agent to build the schema from it. It reads the document, proposes the entities, parts and enums, and creates them once you approve.
Ask for the plan first
Before anything is created, ask the agent to summarise what it found and what it will create. Then check the result in Schema before you build data on it: renaming or removing fields once records exist takes a migration.
Next steps
- Building a Game Schema — review and apply the changes an agent makes
- Working with Game Data — the records an agent creates
- Access Control — what the game can do with them
Last updated on 16 Oct 2026