Agents & MCP
vastframe is built to be driven by AI coding agents. Its MCP server lets an agent such as Claude Code or GitHub Copilot do what a developer does in the console: create a project, deploy a folder, read logs, design the database, share the app and ask for integration access.
https://mcp.<your-domain>/mcp
The server speaks MCP over Streamable HTTP. Every tool is a thin wrapper over the same operations as the REST API and the CLI, so permissions, approvals and the audit log behave identically whichever you use.
Two MCP surfaces
| Surface | Endpoint | Who uses it | Tools |
|---|---|---|---|
| Platform MCP | https://mcp.<your-domain>/mcp | Coding agents working for a person | Projects, deploys, logs, database, storage, access, settings, integrations, docs. See the tools reference. |
| Gateway MCP (per project) | https://<project>.apps.<your-domain>/_tc/int/_mcp | Agents running inside an app | Only the integrations the project is granted. See Gateway MCP. |
How an agent builds an app
- Calls
get_docsonce to learn the SDK. - Writes the code, then calls
deploywith the files and, if the code uses integrations that need approval, a one-linepurpose. - Reads the result: the live URL, a preview URL, the build log tail and any access requests created.
- Uses
describe_schema,apply_migrationandrun_sqlfor the database, andset_accessto share the app.
The app goes live even while access requests are pending; ungranted calls answer ACCESS_PENDING until someone approves.
In this section
- Connect your client: Claude Code, Claude Desktop, claude.ai, GitHub Copilot, Cursor and other clients
- Authentication: dev tokens today, OAuth coming
- Tools reference: Build and deploy, Data and storage, Access, settings and integrations
- Gateway MCP: the per-project integration surface
- Example prompts
Was this page helpful?