Local / Compatible API — any OpenAI-compatible server (Ollama, vLLM, LM Studio, LiteLLM …)
MCP servers — user-scope (~/.claude.json) and project-scope (.mcp.json) with full add / edit / delete
Allowed directories — file-system paths Claude Code may access
Model tiers — Sonnet (primary), Haiku (small/fast), Opus (complex), with live model discovery
All settings are written to ~/.claude/settings.json and work seamlessly with both the Claude Code VS Code extension and the claude terminal CLI.
Getting Started
AWS Bedrock
Open the Command Palette → Open Claude Code Bedrock Settings.
On the Provider tab select AWS Bedrock and fill in your AWS profile and region.
On the Models tab pick which model to use for each tier (cross-region inference profiles are recommended).
Click Save Settings.
Local / Compatible API (Ollama, vLLM, LM Studio, …)
Start your local server.
Go to the Provider tab → select Local / Compatible API.
Enter the Base URL — without a trailing /v1 (e.g. http://localhost:11434 for Ollama, http://localhost:8000 for vLLM).
Claude Code appends /v1/messages automatically; the Refresh button fetches /v1/models.
Open the Models tab and click ⟳ Refresh to load the model list, then choose a model for each tier.
Click Save Settings.
Features
Feature
Detail
Provider switching
Toggle between AWS Bedrock and any OpenAI-compatible local server
Live model discovery
Refresh button fetches /v1/models from your local server
MCP server management
Add/edit/delete servers at user or project scope
Dirty indicator
Title bar shows ● when unsaved changes exist
Disable login prompt
Suppress the Anthropic login prompt when using Bedrock or a local server
Requirements
VS Code 1.98 or later
For Bedrock: AWS CLI configured with a named profile (aws configure --profile <name>)
For Local API: a running OpenAI-compatible server
Extension Settings
This extension does not add VS Code settings. All configuration is written to ~/.claude/settings.json.
