VSCodeMCP

One VS Code extension for AI chat, images, video, music, search, web extraction, and account management. Choose from 26 hosted MCP services and use them in VS Code Chat without installing Python or running local servers.

简体中文教程 · Current models and pricing

First image: GPT Image in VS Code

This example uses one low-quality gpt-image-2 request, then queries the same task ID until its image is ready. Screenshots come from the real English VS Code 1.141 interface. Your model, task ID, image URL, and Credits charge will differ.

1. Install the Marketplace extension

Open Ace Data Cloud MCP in the official VS Code Marketplace. Confirm the publisher is AceDataCloud / acedatacloud, choose Install, and reload VS Code if asked. VS Code 1.101 or newer and an Agent chat model with tool calling are required.

2. Get the right API key

New account? Register and sign in first. If no General application is shown, choose Initialize General Application. When creating a separate key, name it for VS Code and review expiration, usage limits and API restrictions before saving.

1. Sign in to Ace Data Cloud → Applications.
2. Open General application. Copy its API key, or choose Manage Keys → Create to make a separate VS Code key. Check OpenAI image-generation access, current price, and balance before submitting a task.
3. If Allowed APIs is enabled, permit both generation and task lookup: /openai/images/generations and /openai/tasks. Copy only the token string, without Bearer or quotation marks. A platform-... management token is for the optional account server, not this image tool.

3. Choose GPT Image tools and authorize

In the Command Palette, run Ace Data Cloud: Choose MCP Services. Select OpenAI for GPT Image; the OpenAI server supplies openai_generate_image and openai_get_task. For this first run, deselect other services. The Ace Data Cloud Account entry uses a different platform token and is not needed here. Choose OK.

Run Ace Data Cloud: Set Shared API Key and paste the application API key. The key is saved in VS Code SecretStorage, not in your workspace. Use Set Service API Key instead when OpenAI should have its own restricted key. If VS Code asks you to trust the MCP server, verify that it connects to https://openai.mcp.acedata.cloud/mcp before accepting.

4. Enable only the two tools needed

Open Chat → Agent and choose a model that supports tool calling. The separate Ace Data Cloud Chat Models extension can provide one; another compatible Chat model works too. In Chat, select Configure Tools. Under OpenAI · Ace Data Cloud, choose Update Tools if its list is empty, then enable openai_generate_image and openai_get_task. You can disable unrelated tools for this example.

5. Submit one generation

Paste this into Chat. You can also copy the no-key prompt file into your workspace's .github/prompts/ folder and run it from Chat.

Use openai_generate_image exactly once. Model: gpt-image-2. Size: 1024x1024. Quality: low. Number of images: 1. Response format: url. Prompt: A single blue paper sphere on a plain cream background, clean studio photograph, no text. VS Code integration test. After submission, report only the task_id and stop. Do not call image generation again or query the task yet.

Before clicking Allow in this Session, inspect the tool input: model=gpt-image-2, size=1024x1024, quality=low, n=1, and response_format=url. The accepted request returns a task_id; it does not mean the image is finished. Save that ID. Do not rerun the generation to check progress.

6. Query that task until it finishes

In Configure Tools, disable openai_generate_image and leave only openai_get_task enabled. Replace TASK_ID in the no-key query prompt with the ID you just received, or paste this into Chat:

Use openai_get_task only for task ID TASK_ID. Do not call openai_generate_image. Tell me whether it has finished and give one clickable link to the image result if available.

If there is no finished_at or result URL yet, wait and query the same ID again. The task is complete when the response has finished_at and response.success=true. A green Chat turn or an accepted task alone is not completion.

7. Open the image and check billing

Open View the generated image. In our verified run, the returned link served a 1024 × 1024 PNG with HTTP 200. Open Ace Data Cloud Usage and match the single generation task ID to its Credits charge; task queries must not create another image charge. Prices and package conversion rates change, so check the current display rather than copying this example's amount.

What you see	What to do
401 or 403	Recheck the application key, service access, expiry, Allowed APIs, and balance. Do not use a platform management token.
Invalid parameters / 400	Copy the exact model, quality, size, and count above.
Pending or no URL	Query the same task ID again. Never resubmit generation just to poll.
429	Wait and reduce concurrency; do not enable automatic paid retries.
5xx, timeout, or failed task	Inspect the original task and usage history before deciding whether to submit another paid request. Share task/trace ID with support, never your key.

Services

The same extension exposes the services below. GPT Image is under OpenAI, and Google Search is under Google Search (serp internally). Each service has its own tools; select it, check its current API permissions and price, and review write confirmations before running it. A service-specific API key can be saved with Set Service API Key. A second copy of the same MCP from an individual extension should be disabled to avoid duplicate tool entries.

Service	Credential
Ace Data Cloud Account	Platform token
AI Chat	API key
Face Transform	API key
Fish Audio	API key
Flux	API key
GLM	API key
Grok Imagine	API key
Hailuo	API key
HappyHorse	API key
Kling	API key
Luma	API key
Maestro	API key
Midjourney	API key
MiniMax	API key
NanoBanana	API key
OpenAI	API key
Producer	API key
Qwen Image	API key
Seedance	API key
Seedream	API key
Google Search	API key
Short URL	API key
Suno	API key
Veo	API key
Wan	API key
Web Extractor	API key

Account management is optional and requires its own platform token from Platform tokens. Run Ace Data Cloud: Set Account Platform Token to save it. API keys are never sent to the account-management server, and platform tokens are never sent to API-service servers.

Credentials and configuration

Credentials are stored in VS Code SecretStorage, not in settings.json. The extension registers hosted MCP definitions; VS Code controls when each selected server starts. Inputs and the saved API key are sent to that server. API requests use your account and current service pricing. Prompt files contain no keys. Use Set Service API Key when one service needs a different key. Saved service keys take precedence over the shared key. Environment variables ACEDATACLOUD_API_TOKEN and ACEDATACLOUD_PLATFORM_TOKEN are used only when the corresponding saved credential is absent.

Clear Saved Credentials removes keys saved by this extension. It does not change environment variables. To stop exposing a service, deselect it with Choose MCP Services. To stop all services, select none.

You can also set acedatacloud.bundle.services in VS Code settings:

{
  "acedatacloud.bundle.services": ["suno", "kling", "qwen-image", "serp"]
}

The new extension ID is acedatacloud.mcp-toolbox; the deleted March bundle ID cannot be reused. Install this extension explicitly, save your key using the command above and remove the old acedatacloud.apiToken plaintext setting. It is no longer read. The retired Sora service is no longer included. Individual Ace Data Cloud service extensions remain available; disable duplicate entries if you also use this bundle.

Development

npm ci
npm test
npm run check
npm run package

The source of truth is MCPs/vscode-bundle. This repository receives reviewed source changes through the MCPs sync pipeline; do not edit its generated service list or source independently.

In MCPs, opt verified services into scripts/mcp_catalog.json using vscode_bundle. Endpoints and credential kinds come from each service's server.json. Run python3 scripts/build_vscode_bundle.py after catalog changes; CI checks the generated list, package settings and README together. A retired service is automatically omitted. Publish runs tests, packages a VSIX and reads back the public Marketplace version.

API documentation · MCP server source · Issues

MIT licensed.