Sorry, your browser does not support inline SVG.

Documentation

The MCP Server

Connect any AI agent directly to your project

The MCP server is the front door to the Stellify development loop: connect your editor, let AI find existing code and wire it into your project, sync back or export. Claude Code, Cursor, VS Code Copilot, Claude Desktop — any MCP-compatible client connects directly to your Stellify project, and instead of reading and rewriting whole files, the agent works with your code the way Stellify stores it: as addressable files, methods, statements, and elements.

This page covers connecting the server and what each tool does. If you want to import an existing codebase, see Importing with MCP.

Setup

  1. Install the server:
npm install -g @stellify/mcp

  1. Generate an API token in User Settings → API Tokens. The token is shown once — store it securely.

  2. Add the server to your client's MCP configuration (e.g. .vscode/mcp.json, Cursor settings, or Claude Desktop's MCP section):

{
  "mcpServers": {
    "stellify": {
      "command": "npx",
      "args": ["@stellify/mcp"],
      "env": {
        "STELLIFY_API_URL": "https://api.stellisoft.com/api/v1",
        "STELLIFY_API_TOKEN": "your-api-token-here"
      }
    }
  }
}

The server operates on your active project — the one currently open in Stellify. Switch projects in the platform and the agent follows.

The Development Loop, Tool by Tool

The server is designed reuse-first: before generating anything new, agents search the shared library and your project for existing code to reference — because referencing a proven unit costs a fraction of regenerating it (Code Reuse explains the model). One pass through the loop looks like:

  1. search_code — find: check whether the feature (or part of it) already exists in the library
  2. reuse_code — wire it in: reference the unit and its dependencies into your project
  3. create_resources / create_file — fill the gaps: scaffold what the library didn't have
  4. create_route — connect it to a URL
  5. run_migration — apply the database changes
  6. run_code / run_tests — prove it: execute and verify
  7. get_assembled_code — inspect the final rendered source
  8. Sync back — commit to your GitHub repo or export; your repo stays the source of truth

Then the next feature, same loop.

Core Tools

These are always available:

Tool Purpose
get_project The active project: name, directories, enabled capabilities
search_code Search the reusable-code index for existing units to reference
search_files / search_methods / search_routes Find code in your project
reuse_code Reference an existing unit (and its dependency closure) into your project
create_resources Scaffold a complete resource stack — Model, Controller, Service, Migration — in one call
create_file Create a single file (controller, model, middleware, class, or Vue component)
create_route / save_route Create or update routes with controller wiring and middleware
run_code Execute a method in your project environment and return the result
run_migration Apply a migration to your connected database
run_tests Run a test file and return per-test results
get_assembled_code Render a file's structured data back into standard source code

Situational Tool Groups

To keep the agent's context lean, less-common tools load on demand via load_tools:

  • editing — surgical statement-level edits and granular inspection: save_file, save_method, create_method, create_statement, add_method_body, replace_method_body, delete_*, get_file, get_method, get_statement, and more
  • frontend — UI and Vue tools: elements, element trees, HTML-to-elements conversion
  • analysis — performance and code-quality audits
  • capabilities — list and enable libraries/packages for the project
  • settings — read and write project configuration profiles
  • contribute — submit_code, for offering your own code into the shared library

You don't need to manage this yourself — agents load groups as their task requires.

Execution Semantics

run_code executes a single method in a sandboxed environment scoped to your project and database, with a timeout (30s default). Schema-changing operations are blocked here by design.

run_migration is the sanctioned path for schema changes. It applies a migration file to your connected database — use it after scaffolding with create_resources.

run_tests runs every test in a test file and reports pass/fail, failure messages, and duration per test. Test runs are wrapped in a database transaction that is rolled back afterwards, so tests never mutate your project data. Test files export to tests/Feature in your Laravel project like any other code.

The Embedded Chat

You don't have to bring your own agent. The editor's built-in AI chat panel uses the same tool surface — describe what you want and it plans tasks, scaffolds resources, edits methods, and builds UI, with changes appearing live in the editor.

Next Steps

Next
Code Reuse