Add Metabase AI Tools

[vimtor]

Description

As can be seen in the screencast, the first question I ask uses an existing query stored in Metabase. The second one is not stored, and the AI is able to figure out the PostgreSQL query.

I plan for future contributions to add more DDL queries besides PostgreSQL to improve accuracy.

Screencast

https://share.cleanshot.com/X7QXDL1Z

Checklist

• I read the extension guidelines
• I read the documentation about publishing
• I ran npm run build and tested this distribution build in Raycast
• I checked that files in the assets folder are used by the extension itself
• I checked that assets used in the README are located outside the metadata folder if they were not generated with our metadata tool

[raycastbot]

Thank you for the update! 🎉

Due to our current reduced availability, the initial review may take up to 10-15 business days

[greptile-apps]

PR Summary

This PR adds AI tools to the Metabase extension, enabling natural language querying of databases and saved questions through Raycast's AI interface.

• The tools section in package.json needs evals in the ai section - please add more test cases following the evals documentation
• API calls in api.ts should handle failed responses with proper error handling before JSON parsing
• launchCommand in run-query.ts and run-question.ts should be wrapped in try/catch blocks
• Consider using showFailureToast from @raycast/utils instead of showToast for error handling in the API calls
• Since there are multiple commands, consider adding "Metabase" as subtitle to each command for better context

_{💡 (1/5) You can manually trigger the bot by mentioning @greptileai in a comment!}

_{8 file(s) reviewed, 11 comment(s)
Edit PR Review Bot Settings | Greptile}

[greptile-apps]

logic: No error handling for failed requests. Should check res.ok before parsing JSON and handle network errors in the catch block. Consider using showFailureToast from @raycast/utils for error handling.

Suggested change

	export async function runQuery(input: { query: string; databaseId: number }) {
	const endpoint = new URL(`/api/dataset`, preferences.instanceUrl);
	return fetch(endpoint.toString(), {
	method: "POST",
	headers: {
	accept: "application/json",
	"content-type": "application/json",
	"x-api-key": preferences.apiToken,
	},
	body: JSON.stringify({
	database: input.databaseId,
	type: "native",
	native: {
	query: input.query,
	},
	}),
	})
	.then((res) => res.json() as Promise<{ data: Query }>)
	.then((res) => res.data);
	}
	export async function runQuery(input: { query: string; databaseId: number }) {
	const endpoint = new URL(`/api/dataset`, preferences.instanceUrl);
	try {
	const response = await fetch(endpoint.toString(), {
	method: "POST",
	headers: {
	accept: "application/json",
	"content-type": "application/json",
	"x-api-key": preferences.apiToken,
	},
	body: JSON.stringify({
	database: input.databaseId,
	type: "native",
	native: {
	query: input.query,
	},
	}),
	});
	if (!response.ok) {
	throw new Error(`HTTP error! status: ${response.status}`);
	}
	const json = await response.json() as { data: Query };
	return json.data;
	} catch (error) {
	showFailureToast(error instanceof Error ? error.message : "Failed to run query");
	throw error;
	}
	}

[greptile-apps]

syntax: Header name should be 'content-type' (hyphenated) instead of 'contentType' to match other requests and HTTP standards

Suggested change

	accept: "application/json",
	contentType: "application/json",
	"x-api-key": preferences.apiToken,
	accept: "application/json",
	"content-type": "application/json",
	"x-api-key": preferences.apiToken,

[greptile-apps]

logic: Evals are empty arrays. Need at least one eval with expected output to validate AI behavior. See https://developers.raycast.com/ai/write-evals-for-your-ai-extension

[greptile-apps]

logic: getQuestions() call should be wrapped in try/catch to handle API errors gracefully

Suggested change

	export default async function () {
	const questions = await getQuestions();
	export default async function () {
	try {
	const questions = await getQuestions();

[greptile-apps]

style: Function lacks TypeScript return type annotation

[greptile-apps]

style: runQuery should be wrapped in try/catch with showFailureToast from @raycast/utils for better error handling

Suggested change

	try {
	const ddl = await runQuery({
	query: ddlQueries[database.engine],
	databaseId: database.id,
	});
	schema.tables = JSON.parse(ddl.rows[0][0]);
	if (ddl.rows[0][1]) {
	schema.enums = JSON.parse(ddl.rows[0][1]);
	}
	} catch {
	try {
	const ddl = await runQuery({
	query: ddlQueries[database.engine],
	databaseId: database.id,
	});
	schema.tables = JSON.parse(ddl.rows[0][0]);
	if (ddl.rows[0][1]) {
	schema.enums = JSON.parse(ddl.rows[0][1]);
	}
	} catch (error) {
	showFailureToast("Failed to fetch database schema", error);

[greptile-apps]

logic: runQuery should be wrapped in try/catch to handle potential API errors gracefully

Suggested change

	export default async function (input: Input) {
	const result = await runQuery(input);
	return { result };
	export default async function (input: Input) {
	try {
	const result = await runQuery(input);
	return { result };
	} catch (error) {
	throw new Error(`Failed to run query: ${error.message}`);
	}

[greptile-apps]

logic: getDatabase call should be wrapped in try/catch to handle API errors

Suggested change

	export const confirmation: Tool.Confirmation<Input> = async (input) => {
	const database = await getDatabase(input.databaseId);
	export const confirmation: Tool.Confirmation<Input> = async (input) => {
	try {
	const database = await getDatabase(input.databaseId);

[greptile-apps]

logic: runQuestion should be wrapped in try/catch to handle potential API errors gracefully

Suggested change

	export default async function (input: Input) {
	const result = await runQuestion(input);
	return { result };
	}
	export default async function (input: Input) {
	try {
	const result = await runQuestion(input);
	return { result };
	} catch (error) {
	throw new Error(`Failed to run question: ${error.message}`);
	}
	}

[greptile-apps]

logic: question.name could be undefined - needs null check like other fields

[vimtor]

I'm not sure if I should do the AI changes pointed out @pernielsentikaer