MCP Zero-Trust Proxy
A drop-in reverse proxy that adds zero-trust security to any MCP server — no code changes required.
docker run -e MCP_TARGET=localhost:3000 -e AUTH_PROVIDER=github -p 8080:8080 \
ghcr.io/anoblescm/mcp-zero-trust-proxy
接入你的智能体
從發布者安裝。不是 PlugMCP 端點。
遠端 URL 還是本機指令? 兩種安裝方式的差異
來自發布者
A drop-in reverse proxy that adds zero-trust security to any MCP server — no code changes required.
docker run -e MCP_TARGET=localhost:3000 -e AUTH_PROVIDER=github -p 8080:8080 \
ghcr.io/anoblescm/mcp-zero-trust-proxy
Open-source zero-trust proxy for MCP servers — adds OAuth 2.1, tool-level RBAC, audit logging, and rate limiting to any MCP server. One Docker command, zero code changes.
MCP (Model Context Protocol) is how AI agents connect to tools — Claude, Cursor, Copilot all use it. But authentication is optional in the spec. This proxy sits between your MCP clients and servers to enforce security:
mcp-remote, the most popular OAuth workaround# 1. Pull
docker pull ghcr.io/anoblescm/mcp-zero-trust-proxy:latest
# 2. Create config.yaml
cat > config.yaml <<EOF
server:
upstream_url: "http://host.docker.internal:3000"
listen_addr: ":8080"
auth:
provider: "github"
client_id: "your-github-client-id"
client_secret: "\${OAUTH_CLIENT_SECRET}"
redirect_url: "http://localhost:8080/auth/callback"
roles:
- name: "admin"
allowed_tools: []
- name: "readonly"
allowed_tools: []
audit:
enabled: true
output: "stdout"
EOF
# 3. Run
docker run -p 8080:8080 \
-v ./config.yaml:/etc/mcpproxy/config.yaml \
-e OAUTH_CLIENT_SECRET=your_secret \
ghcr.io/anoblescm/mcp-zero-trust-proxy:latest \
--config /etc/mcpproxy/config.yaml
# 4. Verify
curl http://localhost:8080/health
# {"status":"ok"}
go install github.com/AnobleSCM/mcp-zero-trust-proxy/cmd/mcpproxy@latest
mcpproxy --config ./config.yaml
Or download a release binary from the releases page.
AI Client (Claude, Cursor, Copilot)
│
▼
┌─────────────────────────┐
│ MCP Zero-Trust Proxy │
│ │
│ 1. Body size check │
│ 2. Auth (OAuth 2.1) │
│ 3. Rate limit │
│ 4. JSON-RPC parse │
│ 5. RBAC check │
│ 6. Forward to upstream │
│ 7. Filter tools/list │
│ 8. Audit log │
└─────────────────────────┘
│
▼
Your MCP Server
(unchanged)
All config is in a single YAML file. See configs/example.yaml for the full reference.
RBAC example:
roles:
- name: "admin"
allowed_tools: [] # empty = all tools
- name: "readonly"
allowed_tools: ["tools/list", "resources/read"]
- name: "restricted"
allowed_tools: ["read_file", "search_files"]
deny_tools: ["delete_file"]
user_roles:
mapping:
"[email protected]": "admin"
"[email protected]": "readonly"
default: "readonly"
Single binary, ~10MB. Sub-millisecond proxy overhead:
Run benchmarks: ./scripts/benchmark.sh
236+ tests across 9 packages:
go test ./...
go test -race ./... # zero data races
PRs welcome. The codebase is standard Go — no frameworks, minimal dependencies.
go test ./... to verifygo test ./... againFound a vulnerability? See SECURITY.md for responsible disclosure.
Open-source zero-trust proxy for MCP servers — adds OAuth 2.1, tool-level RBAC, audit logging, and rate limiting to any MCP server. One Docker command, zero code changes.
工具
尚無工具清單。遠端伺服器會在首次健康檢查後取得工具清單。
常見問題
在本頁複製適用於 Cursor、Claude Desktop、Claude Code 或 VS Code 的 stdio 設定。它使用發布者的安裝方式,而不是 PlugMCP 代理伺服器。
這筆記錄未列出必要的環境變數。
MCP Zero Trust Proxy 透過 stdio 在本機執行。
上次檢查時間:2026-10-28。 Open-source zero-trust proxy for MCP servers — adds OAuth 2.1, tool-level RBAC, audit logging, and rate limiting to any MCP server. One Docker command, zero code changes.