penpot/mcp/packages/server/src/tools/ExecuteCodeTool.ts
Dominik Jain 7a52550889 Make clear that ExecuteCodeTool serialises automatically
LLMs sometimes decide to apply serialisation themselves, which is unnecessary,
and which this seeks to prevent.
2026-02-15 22:20:38 +01:00

79 lines
3.5 KiB
TypeScript

import { z } from "zod";
import { Tool } from "../Tool";
import type { ToolResponse } from "../ToolResponse";
import { TextResponse } from "../ToolResponse";
import "reflect-metadata";
import { PenpotMcpServer } from "../PenpotMcpServer";
import { ExecuteCodePluginTask } from "../tasks/ExecuteCodePluginTask";
import { ExecuteCodeTaskParams } from "@penpot/mcp-common";
/**
* Arguments class for ExecuteCodeTool
*/
export class ExecuteCodeArgs {
static schema = {
code: z
.string()
.min(1, "Code cannot be empty")
.describe("The JavaScript code to execute in the plugin context."),
};
/**
* The JavaScript code to execute in the plugin context.
*/
code!: string;
}
/**
* Tool for executing JavaScript code in the Penpot plugin context
*/
export class ExecuteCodeTool extends Tool<ExecuteCodeArgs> {
/**
* Creates a new ExecuteCode tool instance.
*
* @param mcpServer - The MCP server instance
*/
constructor(mcpServer: PenpotMcpServer) {
super(mcpServer, ExecuteCodeArgs.schema);
}
public getToolName(): string {
return "execute_code";
}
public getToolDescription(): string {
return (
"Executes JavaScript code in the Penpot plugin context.\n" +
"IMPORTANT: Before using this tool, make sure you have read the 'Penpot High-Level Overview' and know " +
"which Penpot API functionality is necessary and how to use it.\n" +
"You have access two main objects: `penpot` (the Penpot API, of type `Penpot`), `penpotUtils`, " +
"and `storage`.\n" +
"`storage` is an object in which arbitrary data can be stored, simply by adding a new attribute; " +
"stored attributes can be referenced in future calls to this tool, so any intermediate results that " +
"could come in handy later should be stored in `storage` instead of just a fleeting variable; " +
"you can also store functions and thus build up a library).\n" +
"Think of the code being executed as the body of a function: " +
"The tool call returns whatever you return in the applicable `return` statement, if any. " +
"You can return arbitrary JS objects; no need to apply JSON.stringify.\n" +
"If an exception occurs, the exception's message will be returned to you.\n" +
"Any output that you generate via the `console` object will be returned to you separately; so you may use it " +
"to track what your code is doing, but you should *only* do so only if there is an ACTUAL NEED for this! " +
"VERY IMPORTANT: Don't use logging prematurely! NEVER log the data you are returning, as you will otherwise receive it twice!\n" +
"VERY IMPORTANT: In general, try a simple approach first, and only if it fails, try more complex code that involves " +
"handling different cases (in particular error cases) and that applies logging."
);
}
protected async executeCore(args: ExecuteCodeArgs): Promise<ToolResponse> {
const taskParams: ExecuteCodeTaskParams = { code: args.code };
const task = new ExecuteCodePluginTask(taskParams);
const result = await this.mcpServer.pluginBridge.executePluginTask(task);
if (result.data !== undefined) {
return new TextResponse(JSON.stringify(result.data, null, 2));
} else {
return new TextResponse("Code executed successfully with no return value.");
}
}
}