AgentCore MCP Proxy
Bridges AI IDEs with AWS Bedrock AgentCore Gateway using Cognito OAuth and hardware-bound token encryption for secure, authenticated access to dynamically discovered tools.
README
AgentCore MCP Proxy
A generic MCP (Model Context Protocol) proxy server that bridges AI IDEs (Kiro, Claude Desktop, etc.) with any AWS Bedrock AgentCore Gateway. It provides secure, authenticated access to tools exposed by AgentCore endpoints using Cognito OAuth authentication and hardware-bound token encryption.
Multiple AgentCore gateways can be configured simultaneously — each gets isolated token storage so they never interfere with each other.
Architecture
┌─────────┐ stdio ┌──────────┐ HTTPS ┌──────────────────┐
│ Kiro │◀──────────────────▶│ MCP │◀────────────────▶│ AgentCore │
│ IDE │ │ Proxy │ │ Gateway │
└─────────┘ └──────────┘ └──────────────────┘
│ │
│ │
┌────┴────┐ ┌───────┴────────┐
│ Cognito │ │ Your Backend │
│ OAuth │ │ APIs │
└─────────┘ └────────────────┘
- Kiro IDE communicates with the MCP proxy over stdio
- Proxy authenticates via Cognito OAuth (browser-based)
- Authenticated requests are forwarded to AgentCore Gateway over HTTPS
- Tools are dynamically discovered from the gateway (no hardcoding)
Key Components
| File | Purpose |
|---|---|
server.py |
MCP server with stdio transport, tool discovery, and error handling |
auth_manager.py |
Dual auth (Federate + Cognito User), OAuth flow, token caching |
token_encryption.py |
Hardware-bound AES-256 encryption using machine UUID |
gateway_client.py |
Strands MCPClient integration with Bearer token auth |
config.py |
Environment-based configuration with validation |
Features
- 🔐 Dual Authentication: Amazon Corporate SSO (Federate) or Cognito User Pool
- 🔄 Automatic Token Refresh: 30-day validity with transparent refresh
- 🔒 OS Keyring Integration: Encryption keys stored in macOS Keychain, Windows Credential Locker, or Linux Secret Service (file-based fallback for headless environments)
- 🔑 Hardware-Bound Encryption: Tokens encrypted with machine-specific UUID (non-portable)
- 🌐 Multi-Gateway Support: Configure multiple AgentCore endpoints simultaneously with isolated token storage per gateway
- 🛠️ Dynamic Tool Discovery: Tools auto-discovered from AgentCore Gateway
- 🖥️ Cross-Platform: macOS, Linux, and Windows
- 📦 Easy Installation:
uvxorpip
Available Tools
Tools are dynamically discovered from the AgentCore Gateway at startup — no hardcoding required. Whatever tools your gateway exposes will automatically appear in your IDE.
Additionally, if AMAZON_IDP_NAME is configured, two authentication management tools are exposed:
| Tool | Description |
|---|---|
authenticate_with_federate |
Initiate Amazon Corporate SSO authentication |
authenticate_with_cognito_user |
Initiate Cognito User Pool authentication |
Security Overview
The server implements multi-layered security:
- OAuth 2.0 Authorization Code Flow with AWS Cognito
- Encryption keys stored in OS keyring (macOS Keychain / Windows Credential Locker / Linux Secret Service) with file-based fallback
- Hardware-bound token encryption (AES-256, non-portable across machines)
- Per-gateway token isolation — each
AGENTCORE_GATEWAY_URLgets its own encrypted token store - ID Token used for gateway authorization (contains user identity claims)
- Automatic token refresh with 30-day validity
Token storage layout:
~/.agentcore_mcp_proxy/
├── <hash-of-gateway-url-1>/
│ └── tokens.enc
├── <hash-of-gateway-url-2>/
│ └── tokens.enc
For the full security architecture, threat model, and cryptographic details, see SECURITY.md.
Kiro IDE Configuration
Add to ~/.kiro/settings/mcp.json:
{
"mcpServers": {
"agentcore_mcp_proxy": {
"command": "uvx",
"args": [
"--from",
"git+https://github.com/sameerbattoo/awsagentcore_proxy_mcp_server.git",
"agentcore_mcp_proxy"
],
"env": {
"AGENTCORE_GATEWAY_URL": "https://your-gateway.gateway.bedrock-agentcore.us-west-2.amazonaws.com/mcp",
"COGNITO_USER_POOL_ID": "us-west-2_XXXXXXXXX",
"COGNITO_CLIENT_ID": "your_client_id_here",
"COGNITO_DOMAIN": "your-domain.auth.us-west-2.amazoncognito.com",
"TOKEN_TYPE": "id_token"
},
"disabled": false,
"autoApprove": []
}
}
}
Multiple Gateways
You can configure multiple AgentCore endpoints simultaneously. Each entry gets its own isolated token storage (keyed by a SHA-256 hash of the gateway URL), so they never interfere with each other:
{
"mcpServers": {
"gateway_a": {
"command": "uvx",
"args": [
"--from",
"git+https://github.com/sameerbattoo/awsagentcore_proxy_mcp_server.git",
"agentcore_mcp_proxy"
],
"env": {
"AGENTCORE_GATEWAY_URL": "https://gateway-a.gateway.bedrock-agentcore.us-west-2.amazonaws.com/mcp",
"COGNITO_USER_POOL_ID": "us-west-2_AAAAAAA",
"COGNITO_CLIENT_ID": "client_id_a",
"COGNITO_DOMAIN": "domain-a.auth.us-west-2.amazoncognito.com",
"TOKEN_TYPE": "id_token"
}
},
"gateway_b": {
"command": "uvx",
"args": [
"--from",
"git+https://github.com/sameerbattoo/awsagentcore_proxy_mcp_server.git",
"agentcore_mcp_proxy"
],
"env": {
"AGENTCORE_GATEWAY_URL": "https://gateway-b.gateway.bedrock-agentcore.us-east-1.amazonaws.com/mcp",
"COGNITO_USER_POOL_ID": "us-east-1_BBBBBBB",
"COGNITO_CLIENT_ID": "client_id_b",
"COGNITO_DOMAIN": "domain-b.auth.us-east-1.amazoncognito.com",
"TOKEN_TYPE": "id_token"
}
}
}
}
### Environment Variables
All configuration is via environment variables (set in `env` block above or via `.env` file):
| Variable | Required | Description |
|----------|----------|-------------|
| `AGENTCORE_GATEWAY_URL` | Yes | AgentCore Gateway endpoint URL |
| `COGNITO_USER_POOL_ID` | Yes | Cognito User Pool ID |
| `COGNITO_CLIENT_ID` | Yes | Cognito App Client ID |
| `COGNITO_DOMAIN` | Yes | Cognito Domain (without `https://`) |
| `TOKEN_TYPE` | No | Token sent to gateway: `access_token` (default) or `id_token` |
| `AMAZON_IDP_NAME` | No | Identity provider name for Federate SSO (omit to skip Federate) |
| `PROXY_HOST` | No | Callback host (default: `localhost`) |
| `PROXY_PORT` | No | Callback port (default: `8080`) |
See `.env.sample` for a template.
#### TOKEN_TYPE
Controls which Cognito token is sent to the AgentCore Gateway as the Bearer token. Defaults to `access_token`. Set to `id_token` if your gateway needs user identity claims (e.g. `cognito:username`, `email`) for authorization or territory-based access control.
#### AMAZON_IDP_NAME
When set, enables dual authentication mode: users can choose between Amazon Corporate SSO (Federate/Midway) and Cognito User Pool login. The value should match the identity provider name configured in your Cognito User Pool (e.g. `Amazon`, `federate`). When omitted, the server skips Federate entirely and goes straight to Cognito User Pool authentication.
### First-Time Setup
1. Add the MCP configuration to `~/.kiro/settings/mcp.json`
2. Restart Kiro IDE
3. On first tool call, a browser window opens for Cognito authentication
4. Complete login (Amazon SSO or username/password)
5. Tokens are cached encrypted for 30 days with automatic refresh
### Verifying Connection
After configuration, the MCP server appears in Kiro's MCP panel. Verify by asking Kiro to "list my territories" or "list customers".
---
## Building and Testing Locally
### Prerequisites
- Python 3.10+
- `uv` package manager ([install guide](https://docs.astral.sh/uv/getting-started/installation/))
- `npx` (for MCP Inspector testing)
### Quick Start
```bash
# Clone and setup
git clone <repo-url>
cd proxy_mcp_server
uv venv
source .venv/bin/activate
# Install in development mode
uv pip install -e .
# Configure environment
cp .env.sample .env
# Edit .env with your values
# Run the server
agentcore_mcp_proxy
Build, Test & Deploy Script
The repository includes Build_Test_Deploy.sh which automates the full pipeline:
chmod +x Build_Test_Deploy.sh
./Build_Test_Deploy.sh
The script performs these steps:
- Cleans previous builds
- Builds the package with
uv build - Installs in editable mode for testing
- Verifies the installation (import check + entry point)
- Opens MCP Inspector for interactive testing (
http://localhost:6274) - Guides you through Kiro testing
- Deploys to TestPyPI or PyPI
Manual Build Steps
# Clean
rm -rf dist/ build/ *.egg-info
# Build
uv build
# Install locally
uv pip install -e . --force-reinstall --no-deps
# Verify
python -c "from agentcore_mcp_proxy import main; print('OK')"
which agentcore_mcp_proxy
Testing with MCP Inspector
npx @modelcontextprotocol/inspector $(which agentcore_mcp_proxy)
Opens a browser at http://localhost:6274 where you can interactively test tools.
Publishing to PyPI
# TestPyPI (recommended first)
uv publish --publish-url https://test.pypi.org/legacy/
# Production PyPI
uv publish
After publishing, users can install with:
uvx agentcore_mcp_proxy@latest
# or
pip install agentcore-mcp-proxy
Troubleshooting
Port 8080 Already in Use
The server automatically kills blocking processes. If that fails:
lsof -ti :8080 | xargs kill -9
"Encryption key mismatch"
Tokens were copied from another machine or hardware changed:
rm -rf ~/.agentcore_mcp_proxy/
# Re-authenticate on next use
"Missing required environment variables"
Ensure all required env vars are set. See Environment Variables above.
No Midway Session (Federate only)
mwinit
# Or use Cognito User authentication instead
Gateway 401 Unauthorized
Delete cached tokens and re-authenticate:
rm -rf ~/.agentcore_mcp_proxy/
License
MIT License — See LICENSE file.
Recommended Servers
playwright-mcp
A Model Context Protocol server that enables LLMs to interact with web pages through structured accessibility snapshots without requiring vision models or screenshots.
Magic Component Platform (MCP)
An AI-powered tool that generates modern UI components from natural language descriptions, integrating with popular IDEs to streamline UI development workflow.
Audiense Insights MCP Server
Enables interaction with Audiense Insights accounts via the Model Context Protocol, facilitating the extraction and analysis of marketing insights and audience data including demographics, behavior, and influencer engagement.
VeyraX MCP
Single MCP tool to connect all your favorite tools: Gmail, Calendar and 40 more.
graphlit-mcp-server
The Model Context Protocol (MCP) Server enables integration between MCP clients and the Graphlit service. Ingest anything from Slack to Gmail to podcast feeds, in addition to web crawling, into a Graphlit project - and then retrieve relevant contents from the MCP client.
Kagi MCP Server
An MCP server that integrates Kagi search capabilities with Claude AI, enabling Claude to perform real-time web searches when answering questions that require up-to-date information.
E2B
Using MCP to run code via e2b.
Neon Database
MCP server for interacting with Neon Management API and databases
Exa Search
A Model Context Protocol (MCP) server lets AI assistants like Claude use the Exa AI Search API for web searches. This setup allows AI models to get real-time web information in a safe and controlled way.
Qdrant Server
This repository is an example of how to create a MCP server for Qdrant, a vector search engine.