1. Conduid
  2. Cloud
  3. Bedrock Claude Code
MCP server · Cloud

Bedrock Claude Code

VS Code settings panel for Claude Code on AWS Bedrock and local LLMs

Unclaimed cloud
34Low

Scored 6 months ago · breakdown

About Bedrock Claude Code

Bedrock Claude Code is an MCP server in the Cloud category: vS Code settings panel for Claude Code on AWS Bedrock and local LLMs.

Install

From source
git clone https://github.com/easytocloud/bedrock-claude-code

This server has no ConduID identity, so agent calls to it are not receipted. Pin the version you install and review the source before granting it credentials.

Ask AI

✦

Ask AI about Bedrock Claude Code

Powered by Claude · Grounded in docs

I know everything about Bedrock Claude Code. Ask me about installation, configuration, usage, or troubleshooting.

Security checks

  • ✓README presentThe repository ships documentation.

5 checks have not been run yet.

README

Claude Code Personae

A VS Code extension for managing Claude Code configurations through composable, reusable presets.

Claude Code Personae screenshot

Concept

Instead of editing JSON files by hand, this extension lets you build presets from three types of building blocks:

  • Providers — API backends: Anthropic Direct, AWS Bedrock, or any Anthropic-compatible proxy (LiteLLM, OpenRouter, Ollama, vLLM, LM Studio, …)
  • MCP Server Groups — named collections of MCP servers
  • Directory Groups — additional directories Claude Code may access

A preset bundles one provider + any number of MCP server groups + any number of directory groups into a single switchable configuration.

Presets are then assigned to scopes:

Scope Config files written Dropdown options
Global ~/.claude/settings.json, ~/.claude.json Any preset, or Manual
Workspace {workspace}/.claude/settings.json, .mcp.json Any preset, Inherit from Global, or Manual

Switching a scope's preset instantly reconfigures Claude Code — no manual file editing required.

Storage

All building blocks, presets, and scope assignments are stored in a single file:

~/.claude/coder-profiles.json

On Save All, the extension resolves the active presets and writes the resulting flat configuration into Claude Code's own files. When a workspace scope is set to Inherit from Global, any previously written workspace-level settings are cleaned up automatically.

A sample configuration is included in examples/coder-profiles.json.

Getting Started

  1. Open the Command Palette → Open Claude Code Settings
  2. Create a Provider (e.g. AWS Bedrock with your profile and region)
  3. Optionally create MCP Server Groups and Directory Groups
  4. Create a Preset that combines your provider with any groups
  5. Assign the preset to the Global or Workspace scope
  6. Click Save All

AWS Bedrock

  • Select provider type Bedrock, fill in your AWS profile name and region
  • Pick models from the smart presets or click Fetch models from AWS to discover all inference profiles and foundation models in your account
  • Test models — each model slot has a Test pill that calls aws bedrock-runtime converse to verify profile and model access. Results (OK / Fail) are persisted on the provider
  • Optionally set an auth-refresh command (e.g. aws sso login --profile my-profile)
  • Claude Code's login/logout commands are automatically disabled when using Bedrock
  • If $AWS_CONFIG_FILE is set or ~/.aws/config is a symlink, the resolved config path is shown as AWS Config (read-only). If you use easytocloud aws-envs, an AWS Env dropdown appears instead — each provider stores its own env selection in coder-profiles.json, so different providers can point to different AWS environments independently

Local / Other (Ollama, vLLM, LM Studio, LiteLLM, OpenRouter, …)

  • Select provider type Local / Other, enter the base URL (e.g. http://localhost:11434)
  • Click Fetch available models to discover models from /v1/models
    • Single-model endpoints auto-select the model for all tiers
    • Multi-model endpoints show a selection-only dropdown (sorted alphabetically, with slot-matching models listed first)
    • If the endpoint does not expose /v1/models, the fetch fails gracefully and you can type model IDs manually
  • Test models — each model slot has a Test pill that sends a minimal request to POST /v1/messages to verify Anthropic API compatibility. Results (OK / Fail) are persisted on the provider
  • Credential — enter an API key or token directly, or use an op://Vault/Item/field 1Password reference. A compact pill toggle switches between API Key (x-api-key header) and Token (Authorization: Bearer). The value is preserved when toggling
  • Standalone mode is on by default — it blocks all traffic to Anthropic (telemetry, updates, login prompt). Disable it only if your proxy forwards requests to Anthropic directly and you need the Anthropic login flow.

OpenRouter

  • Select provider type Local / Other, enter https://openrouter.ai as the base URL (the /api suffix is added automatically; /v1 is stripped if pasted)
  • The auth pill auto-switches to Token when an openrouter.ai URL is detected. Paste your OpenRouter API key or use an op:// reference for 1Password
  • Click Test on each model slot to verify compatibility before saving

Anthropic Direct

  • The built-in Anthropic provider is always available
  • Optionally set your API key in the provider editor (enter a key from console.anthropic.com, or use an op:// reference for 1Password)
  • Without an API key, you'll need to run /login with an Anthropic Max or Pro plan
  • Model selection is not shown — Claude Code uses its built-in defaults (Sonnet for primary, Haiku for small/fast tasks, Opus for complex tasks)

Features

Feature Detail
Composable presets Mix and match providers, MCP servers, and directories
Scope management Global and per-workspace configurations with inheritance
Provider types Anthropic Direct, AWS Bedrock, Local / Other (any Anthropic-compatible proxy)
MCP server groups Reusable named collections of MCP servers (stdio, HTTP, SSE)
Directory groups Additional directories Claude Code may access
Live model discovery Fetch available models queries /v1/models; auto-selects single-model endpoints
Model compatibility testing Per-slot Test pill verifies the model speaks Anthropic's /v1/messages API; results persisted per provider
1Password support Enter op://Vault/Item/field as the credential — the extension writes apiKeyHelper for Claude Code to resolve at startup
Filterable dropdowns Type-to-filter combobox for AWS profiles (100+) and model lists (500+ OpenRouter) — slot-matching models grouped first, alphabetical sort, keyboard nav, match highlighting
Quick-switch status bar Click the status bar item to switch presets for global or workspace scope without opening the panel
Import / Export Share presets between machines or team members — credentials are scrubbed on export, recipients fill in their own
Draft auto-save Unsaved changes persist across panel close and are restored on re-open
Dirty indicator Title bar shows ● when unsaved changes exist
Drawer-based editing Slide-out panels for editing all building blocks
Inherit mode Workspace scope can inherit from global — cleans up workspace files
Login prompt suppression Automatic for Bedrock (always on) and Proxy (on by default, overridable). Sets both env vars in ~/.claude/settings.json and the claudeCode.disableLoginPrompt VS Code setting at the right scope (global or workspace).

Requirements

  • VS Code 1.98 or later
  • For Bedrock: AWS CLI configured with a named profile (aws configure --profile <name>)
  • For Local / Other: a running Anthropic-compatible server (/v1/messages endpoint)
  • For 1Password credentials: 1Password CLI (op) installed and signed in

Extension Settings

This extension does not add VS Code settings. All configuration is managed through ~/.claude/coder-profiles.json and resolved into Claude Code's own files on save.

Release Notes

See CHANGELOG.md.

README copy captured when this server was indexed; it may lag the repository. The original is authoritative.

Questions

About Bedrock Claude Code

How do I install Bedrock Claude Code?

No package has been published for this server, so install it from source: clone the repository and follow its README.

Is Bedrock Claude Code safe to use with an AI agent?

Its trust score is 34 out of 100 (low). It passes 1 of the 1 static security checks run so far. It has no ConduID identity yet, so agent calls to it are not receipted.

Is Bedrock Claude Code still maintained?

Conduid hasn't recorded a commit date for this repository yet. Check the repository directly for recent activity.