Clampd is a runtime firewall for AI agents, LLM applications, and MCP (Model Context Protocol) servers. Every tool call an agent makes - database queries, API requests, file access, shell commands, payment transactions - gets intercepted, classified, scored, and either signed or blocked before it reaches the downstream service. This addresses the core failure mode of agentic systems: a model can be socially engineered, prompt-injected, or simply hallucinate, but the security boundary lives outside the model where it cannot be talked around.
The platform ships with 264 detection rules across 12 tool-call categories, evaluated in microseconds with single-digit-millisecond end-to-end latency on commodity hardware. Capabilities include Ed25519-signed scope tokens that replace long-lived credentials, behavioral anomaly detection that learns normal agent patterns, cross-agent correlation for multi-step attack detection, an emergency kill switch, prompt-injection scanning, and compliance reports for HIPAA, GDPR, SOC 2, and PCI-DSS. Integrations cover OpenAI, Anthropic, LangChain, CrewAI, Google ADK, Claude Code, Cursor, and any MCP server via a transparent proxy. The SDK is one line.
Clampd is self-hosted and source-available - your agent traffic never leaves your infrastructure. Deployment is a single Docker Compose command, production-ready in five minutes. The product surface includes language SDKs, an MCP proxy, the clampd-guard agent for IDE assistants, a CLI with TUI, and a security dashboard with live risk feed, policy editor, and audit trail. Built for security and platform engineering teams shipping AI agents to production where the cost of a bad tool call is real.
