Vertex Anthropic Provider for VS Code

A VS Code extension that registers Anthropic Claude models (via Google Cloud Vertex AI) as native language model chat providers. Once installed, the models appear alongside other models in the VS Code Copilot Chat view — no separate UI, no extra windows.

The extension automatically discovers which Claude models are available in your project and picks the best region, so there’s nothing to configure beyond your GCP Project ID.

How It Works

The extension implements the vscode.LanguageModelChatProvider API to bridge VS Code's built-in Chat UI with Anthropic Claude models hosted on Google Cloud Vertex AI Model Garden. It handles:

Dynamic model discovery — pings candidate models on activation and registers only the ones available in your project.
Auto region detection — tries global first, then regional endpoints (us-east5, europe-west1, asia-southeast1).
Remote model catalog — fetches the candidate model list from a configurable URL so your team can add models without rebuilding.
Interactive Usage Dashboard — Native VS Code Webview dashboard tracking daily costs, cached tokens, and payload diagnostics via Apache ECharts.
API Payload Character Tracking — Automatically computes literal byte sizing across User Text, System rules, Base64 Images, and Tool JSON calls.
Intelligent Prompt Caching — Automatically injects ephemeral caching on systemic boundaries reducing token costs for repeating conversational setups.
Message mapping — translates VS Code chat messages (user, assistant, system, tool results, tool calls) into Anthropic’s API format.
Image / vision support — images pasted into chat are sent to Claude as base64 image content blocks.
Streaming responses — streams tokens back to the Chat UI in real time via the Vertex AI SDK.
Tool calling — supports VS Code's native tool-calling flow, forwarding tool definitions and parsing streamed tool-use responses.
Authentication — delegates to Google Cloud Application Default Credentials (ADC), so there are no API keys to manage.

Installation

From `.vsix` (pre-built)

Download the .vsix file from the repository:
vertex-anthropic-0.0.4.vsix
Install it in VS Code using one of the following methods:

Option A — VS Code UI:
- Open the Extensions view (Ctrl+Shift+X)
- Click the ··· menu (top-right of the Extensions sidebar) → Install from VSIX…
- Select the downloaded .vsix file
Option B — Command line:
```
code --install-extension vertex-anthropic-0.0.4.vsix
```
Reload VS Code when prompted.

From Source

git clone https://github.com/jorsm/vertex-anthropic.git
cd vertex-anthropic
npm install
npm run compile

Then press F5 in VS Code to launch the Extension Development Host with the extension loaded.

Prerequisites

Google Cloud Authentication

The extension uses the @anthropic-ai/vertex-sdk, which authenticates via Application Default Credentials (ADC). Before using the extension, authenticate locally:

gcloud auth application-default login

Your GCP project must have the Vertex AI API enabled and access to Claude models in the Model Garden.

VS Code Version

Requires VS Code 1.110.0 or later (for the languageModelChatProviders contribution point).

Configuration

Open VS Code Settings (Ctrl+,) and search for Vertex Anthropic, or add the following to your settings.json:

Setting	Type	Default	Description
`vertexAnthropic.projectId`	`string`	`""`	Your GCP Project ID (required)
`vertexAnthropic.modelCatalogUrl`	`string`	(GitHub raw)	URL to a remote `models.json` catalog. Set to empty to use only the bundled catalog.

Example:

// settings.json
{
  "vertexAnthropic.projectId": "my-gcp-project-123"
}

The region is auto-detected — no configuration needed.

⚠️ The extension will show a warning on activation if projectId is not set.

Usage

Open the Chat view in VS Code (Copilot icon in the sidebar, or Ctrl+Shift+I).
Click the model picker — all discovered Vertex Claude models will be listed (e.g. "Vertex Claude Opus 4.6", "Vertex Claude Sonnet 4.6").
Type a message and send — responses stream in real time from Vertex AI.
Paste images directly into the chat for vision analysis.

The models support tool calling, so they work with VS Code’s built-in agent tools and any extensions that provide tool definitions.

Commands

Vertex Anthropic: Refresh Models (Ctrl+Shift+P) — re-runs model discovery.

Status Bar & Dashboard

An interactive Cost Tracker exists in native VS Code Status bar updating on the fly when tokens resolve.
Click the Status Bar or Run Vertex Anthropic: Show Usage Dashboard to open the local Cost Tracking Webview filtering all Usage & Payload metrics.

Diagnostics

Open the Output panel and select Vertex Anthropic to see detailed logs: catalog loading, region probing, message mapping, token usage, and stream lifecycle.

Candidate Models

The extension discovers models from a JSON catalog. The default bundled catalog includes:

Model ID	Display Name	Max Input	Max Output	Vision	Tools
`claude-opus-4-6`	Claude Opus 4.6	1M	128K	✅	✅
`claude-sonnet-4-6`	Claude Sonnet 4.6	1M	128K	✅	✅
`claude-haiku-4-5`	Claude Haiku 4.5	200K	64K	✅	✅

Only models that respond to a ping in your project are registered. Update the list by pointing vertexAnthropic.modelCatalogUrl to your own models.json.

Development

npm run watch    # compile TypeScript in watch mode
npm run lint     # run ESLint
npm run test     # run tests

Press F5 to launch the Extension Development Host. Use Ctrl+R inside the host window to reload after code changes.

License

See the LICENSE file for details.

Vertex Anthropic Provider

jorsm