Vedis sits between your AI agents and MCP servers. Detects prompt injection, enforces tool policies, filters secrets. One config change, zero code changes.
Sends MCP tool call
via stdio or SSE
Scans, filters, enforces
Receives clean,
policy-compliant request
20+ heuristic patterns for override attempts, role hijacking, encoding tricks. Compound threat scoring with 3 sensitivity levels — low, medium, high.
YAML-based allowlist and blocklist rules. Glob patterns for tool matching. Fine-grained constraints on arguments, methods, and resources.
PII detection: emails, phone numbers, SSNs, credit cards. Secret scanning: AWS keys, GitHub tokens, Stripe keys, JWTs, PEM certificates.
JSONL and SQLite backends. Indexed queries over full request/response trails. Every decision logged with timestamps and threat details.
Sliding window algorithm with configurable per-minute limits. Returns proper JSON-RPC error responses. Prevents runaway agent loops.
Real-time stats and metrics. Interactive injection scanner test. Audit log viewer with filtering. All served from a built-in HTTP server.
# Install Vedis globally $ npm install -g vedis # Generate default config $ vedis init # Start the security proxy $ vedis proxy --upstream "npx -y @modelcontextprotocol/server-filesystem /tmp"