ThinkWatchProject/ThinkWatch

Enterprise AI bastion host for secure AI API and MCP access, with unified proxying, RBAC, audit logs, rate limiting, and cost tracking across OpenAI, Anthropic, Gemini, and self-hosted LLMs.

What it solves

ThinkWatch is an enterprise-grade secure gateway designed to centralize the governance of AI API calls and Model Context Protocol (MCP) tool invocations. It addresses common organizational challenges such as scattered API keys, lack of visibility into token usage and costs, missing access controls for models and tools, and compliance gaps in audit trails.

How it works

ThinkWatch operates as a unified proxy (Gateway) and a management interface (Console). It intercepts all AI requests and MCP tool calls, providing a single control plane for authentication, authorization, and monitoring.

  • AI Gateway: Acts as a drop-in replacement for various AI SDKs and clients (like Cursor or Claude Code), routing requests to providers like OpenAI, Anthropic, and Google Gemini while handling format conversion and streaming.
  • MCP Gateway: Unlike typical proxies, it propagates the real end-user identity to upstream servers using per-user OAuth tokens or PATs, ensuring that audit logs on the upstream side reflect the actual user rather than a shared service account.
  • Governance Engine: Implements a multi-tier RBAC system, virtual API keys with lifecycle management, and a sophisticated quota system using sliding-window rate limits and natural-period budget caps.

Who it’s for

It is built for organizations and engineering teams that need to secure, audit, and govern AI agent proliferation across their teams, specifically those requiring strict identity propagation for MCP tools and detailed cost attribution.

Highlights

  • Unified AI Proxy: Supports multiple API formats (OpenAI, Anthropic) and providers (Azure, Bedrock, Gemini) on a single port.
  • Identity-Aware MCP: Propagates real user identities to upstream tools via encrypted OAuth/PAT vaults, preventing the "shared service account" problem.
  • Advanced Quota Management: Combines sliding-window rate limits and calendar-aligned budget caps with per-model token weighting.
  • Enterprise Security: Features AES-256-GCM encryption, OIDC/SSO integration, and distroless containers for a minimal attack surface.
  • Deep Observability: Uses ClickHouse for high-performance audit logs and provides Prometheus metrics for real-time monitoring.

Related

  • Project
    agentic-community/mcp-gateway-registry

    A governed control plane and gateway for AI assets, providing centralized discovery, authentication, and audit trails for MCP servers, agents, and skills.

  • Project
    obot-platform/obot

    An open-source platform for organizations to manage, secure, and govern their AI ecosystems, providing centralized control over LLM access, MCP servers, and AI activity auditing.

  • Project
    Azure-Samples/AI-Gateway

    An enterprise-grade AI Gateway framework powered by Azure API Management that provides centralized security, cost control, and observability for managing AI models, tools, and agents.

  • Project
    opentokenz/mcpx

    A secure MCP Runtime gateway that provides AI clients with auditable and persistent access to local workspaces, source code, and system tools.

  • Project
    astaxie/TokenHub

    TokenHub is an enterprise token governance infrastructure that provides centralized model routing, access control, and cost optimization for AI model providers.