WaveGuard
Hook your coding agent into physics-based outlier scanning over metrics, logs, or custom series without building a full ML stack.
Overview
WaveGuard is a MCP server for the Operate phase that scans arbitrary data for outliers via a physics-simulation anomaly detection API over streamable HTTP.
What is this MCP server?
- Remote streamable-http MCP at Modal-hosted endpoint (version 3.3.0 in registry)
- Physics-simulation anomaly detection API usable on arbitrary numeric series
- Designed for outlier scans without training a custom model per dataset
- Companion WaveGuardClient repo for client-side integration patterns
- LFMAnomalyDetection GitHub source for the underlying detection approach
- Registry version 3.3.0
- Transport: streamable-http remote MCP
- Source repos: LFMAnomalyDetection and WaveGuardClient on GitHub
What problem does it solve?
You have metrics or batches piling up and no fast, agent-callable way to ask whether the latest window looks statistically weird.
Who is it for?
Indie builders with live metrics or pipeline outputs who want agent-driven anomaly checks without owning model training infrastructure.
Skip if: Teams that need certified fraud detection, deep root-cause RCA, or fully on-prem air-gapped execution with no remote API.
What do I get? / Deliverables
After you add the remote MCP endpoint, your agent can run outlier scans on demand and surface anomaly signals you can tie into alerts or triage.
- Agent-invokable outlier scan results on submitted data
- Repeatable anomaly checks wired into ops or review workflows
Recommended MCP Servers
Journey fit
Outlier detection is most valuable once something is live and you need to spot drift, spikes, or bad batches in production or pipeline data. Monitoring is the canonical shelf because WaveGuard is framed as scanning data for anomalies rather than one-off exploratory charts.
How it compares
Remote anomaly-detection API via MCP, not a local Grafana dashboard or a general Python sandbox.
Common Questions / FAQ
Who is WaveGuard for?
Solo and small-team builders who ship data-backed products and want their AI coding agent to call a hosted outlier scanner over MCP.
When should I use WaveGuard?
Use it in operate and monitoring workflows when you need to sanity-check recent series, batch exports, or experiment metrics for anomalies.
How do I add WaveGuard to my agent?
Register the streamable-http remote URL from the server manifest (Modal-hosted WaveGuard API) in your MCP client, then invoke the server’s anomaly tools from Claude Code, Cursor, or another MCP-capable agent.