Build with AI agents over MCP

The two servers

Connect Claude Code

claude mcp add --transport http idosgames https://site.idosgames.com/api/v2/mcp
claude mcp add idosgames-registry -- npx -y @idosgames/mcp

Restart the session, run /mcp and sign in to "idosgames" in the browser window that opens. Then ask the agent to load the skill idosgames-getting-started and help you build your first app.

Connect Cursor, Codex or another agent

Add both servers to your MCP config:

{
  "mcpServers": {
    "idosgames": { "url": "https://site.idosgames.com/api/v2/mcp" },
    "idosgames-registry": { "command": "npx", "args": ["-y", "@idosgames/mcp"] }
  }
}

Restart the agent, sign in to "idosgames" in the browser window that opens and start with the skill idosgames-getting-started.

Connect Claude or ChatGPT

Add a custom connector with the address https://site.idosgames.com/api/v2/mcp. On first use the assistant asks you to sign in to iDos Games and pick the apps it may work with. The exact menu path for each assistant is shown on idosgames.com/mcp.

Create a new app

create_title creates an app the same way the dashboard does: it takes a slot of your plan and gets the base systems, a template, a ready starter build and a DEV copy for tests. The new app joins the agent's access right away, so the agent can set it up and publish a build next. It needs its own permission "Create new apps" on the sign-in screen. When the plan has no free slot, the agent gets a link where you can upgrade it.

• "Create a new idle RPG called Moon Farm and add three currencies."

Videos

The video studio template - an app where players make short vertical videos - works with agents too. On the remote server plan_video turns an idea into a ready video written by the studio's AI director: format, scene, characters, lines and captions from the studio library. generate_asset makes the characters, scenes, props and music the video needs and puts each one in its place. Both spend the AI credits of your plan, then your Aspis - ask the agent to check the price with estimate_media_cost first. The answer is the video as data plus a link that opens it in your app's editor, where it is watched, edited and saved as MP4 on the device. publish_asset puts the assets into the Workshop of your app as its official content, so its players can use them too. Reading the studio's formats and library (list_video_formats, search_assets) is free, and so are validate_asset and validate_video on the local server.

• "Write a 30-second dialogue video about a cat who hates Mondays for my video studio app and give me the link."

3D models

generate_three_d makes a 3D model from text, a picture or views, and works on a model you already made: texture, skeleton, animation, a new file format. For a character with a skeleton, ready animations from the platform library (idle, walk, run) cost less than new ones: list_animation_library lists them, and the operation library-animation puts the chosen ones on your model, in place by default. Both spend the AI credits of your plan, then your Aspis - ask the agent to check the price with estimate_media_cost first.

A standing character (T-pose or A-pose) can also get the platform's own skeleton, the operation auto-rig: the joints are found automatically, and on the AI screen of the dashboard you can drag them by hand. Library animations go on top of it, and only they: retarget takes only a skeleton from rig. Our rig and library clips are free up to 3 rigs and 3 clips a day per account; above that a rig costs 5 Aspis and a clip 3 Aspis (on the dashboard you can also pay with ad units earned in the iDos Games app on Android).

• "Make a low-poly knight, give it a skeleton and the idle, walk and run animations from the library."

What to ask

• "Create a store with three coin packs and a daily deal."
• "Generate an icon for the sword item and put it into the store."
• "Use the leaderboard system and show the top 100 by score."
• "Upload my build and deploy it."

The agent works only with the apps you allowed, and publishing posts to social networks needs a separate permission.

Built builds are published the same way as from the AI coder - see publishing.