identity-aware-mcp-server
A production-ready MCP server that authenticates agents via OAuth 2.1 Bearer tokens, validates JWTs with JWKS, enforces tool-level scopes and roles, and logs the full delegation chain.
README
Identity-Aware MCP Server — Companion Project
A production-ready reference implementation of an MCP server that requires OAuth 2.1 Bearer tokens, validates them against a JWKS endpoint, enforces tool-level scopes and roles, and logs every access decision with the full delegation chain.
Companion project for: "Building an Identity-Aware MCP Server" (DZone, June 2026)
What this server does
- Authenticates every agent — no anonymous or shared-key access
- Validates JWTs using stateless JWKS-based validation (no DB lookups)
- Enforces tool-level authorization — each tool declares required scopes and roles
- Maintains a tool allow-list — infrastructure-level enforcement (not prompt-level)
- Logs the full delegation chain — every action traces back to a human identity
- Serves MCP discovery endpoints — clients auto-discover auth configuration
Architecture
Agent (Bearer Token)
│
▼
┌─────────────────────────────┐
│ 1. Token Validation (JWKS) │ ← Is this a valid token?
├─────────────────────────────┤
│ 2. Tool Allow-List Gate │ ← Is this tool allowed at all?
├─────────────────────────────┤
│ 3. Scope + Role Check │ ← Can THIS agent call THIS tool?
├─────────────────────────────┤
│ 4. Tool Handler │ ← Execute the actual operation
├─────────────────────────────┤
│ 5. Audit Log │ ← Who did what, on whose behalf?
└─────────────────────────────┘
Requirements
- Python 3.12+
- An OIDC-compatible identity provider (Auth0 free tier works)
Quick Start
# 1. Clone and set up
cd code
python3 -m venv .venv
source .venv/bin/activate
pip install -r requirements.txt
# 2. Configure (edit with your Auth0/Okta details)
cp .env.example .env
# 3. Run the server
python src/server.py
# 4. In another terminal — run the demo agent
python demo/agent.py --demo
Demo Mode
The --demo flag generates self-signed tokens locally so you can test
the full flow without configuring an OAuth provider:
python demo/agent.py --demo
This runs three simulated agents:
- Alice (developer,
database.read + email.sendscopes) → can read, cannot write - Bob (admin, full scopes) → can do everything
- Carol (contractor, zero scopes) → blocked from everything
Testing
python -m pytest tests/ -v
Expected output:
tests/test_server.py::test_protected_resource_endpoint PASSED
tests/test_server.py::test_authorization_server_endpoint PASSED
tests/test_server.py::test_no_token_returns_401 PASSED
tests/test_server.py::test_invalid_token_returns_401 PASSED
tests/test_server.py::test_expired_token_returns_401 PASSED
tests/test_server.py::test_valid_token_read_customer_succeeds PASSED
tests/test_server.py::test_missing_scope_returns_403 PASSED
tests/test_server.py::test_write_scope_required_for_update PASSED
tests/test_server.py::test_missing_role_returns_403 PASSED
tests/test_server.py::test_admin_role_succeeds PASSED
tests/test_server.py::test_unknown_tool_returns_404 PASSED
tests/test_server.py::test_audit_log_entries PASSED
Key Files
| File | Purpose |
|---|---|
src/server.py |
Main MCP server — configuration, routing, auth enforcement |
src/auth/middleware.py |
JWKS cache, TokenValidator, require_auth decorator |
src/auth/discovery.py |
OAuth 2.1 discovery endpoints (RFC 8414 + RFC 9728) |
src/tools/database.py |
Example tools with scope and role declarations |
src/audit/logger.py |
JSON Lines audit log with delegation-chain tracing |
demo/agent.py |
Demo agent that connects with scoped tokens |
Production Hardening
Before deploying:
- [ ] Use a real OAuth 2.1 provider (Auth0, Okta, Entra ID) — not demo keys
- [ ] Enforce HTTPS only — reject HTTP at the network level
- [ ] Set short token lifetimes (15-60 minutes) with refresh token rotation
- [ ] Sanitize Bearer tokens from request logs
- [ ] Run behind a reverse proxy with rate limiting
- [ ] Use PostgreSQL-backed token storage for multi-instance deployments
Article Reference
- Title: Building an Identity-Aware MCP Server
- Publication: DZone
- Date: June 2026
- Author: Pravin Khandke
推荐服务器
Baidu Map
百度地图核心API现已全面兼容MCP协议,是国内首家兼容MCP协议的地图服务商。
Playwright MCP Server
一个模型上下文协议服务器,它使大型语言模型能够通过结构化的可访问性快照与网页进行交互,而无需视觉模型或屏幕截图。
Magic Component Platform (MCP)
一个由人工智能驱动的工具,可以从自然语言描述生成现代化的用户界面组件,并与流行的集成开发环境(IDE)集成,从而简化用户界面开发流程。
Audiense Insights MCP Server
通过模型上下文协议启用与 Audiense Insights 账户的交互,从而促进营销洞察和受众数据的提取和分析,包括人口统计信息、行为和影响者互动。
VeyraX
一个单一的 MCP 工具,连接你所有喜爱的工具:Gmail、日历以及其他 40 多个工具。
graphlit-mcp-server
模型上下文协议 (MCP) 服务器实现了 MCP 客户端与 Graphlit 服务之间的集成。 除了网络爬取之外,还可以将任何内容(从 Slack 到 Gmail 再到播客订阅源)导入到 Graphlit 项目中,然后从 MCP 客户端检索相关内容。
Kagi MCP Server
一个 MCP 服务器,集成了 Kagi 搜索功能和 Claude AI,使 Claude 能够在回答需要最新信息的问题时执行实时网络搜索。
e2b-mcp-server
使用 MCP 通过 e2b 运行代码。
Neon MCP Server
用于与 Neon 管理 API 和数据库交互的 MCP 服务器
Exa MCP Server
模型上下文协议(MCP)服务器允许像 Claude 这样的 AI 助手使用 Exa AI 搜索 API 进行网络搜索。这种设置允许 AI 模型以安全和受控的方式获取实时的网络信息。