BlueRock adds runtime security monitoring for Python MCP servers

BlueRock has released an open source Python sensor that monitors MCP tool calls, resource access, imports, and session events at runtime with no code changes. According to the project, it wraps the Python process at startup and writes structured NDJSON logs for security teams and AI developers.

BlueRock adds runtime security monitoring for Python MCP servers

BlueRock has released a lightweight runtime security sensor for Python applications, with a focus on Model Context Protocol, or MCP, servers. According to the project, it can monitor tool calls, resource access, session lifecycle events, and module imports without requiring any changes to application code.

The project says the sensor runs by wrapping the Python process before the target script starts. That means events from the application itself, plus its dependencies and transitive dependencies, are captured from the beginning of execution rather than after the program is already running.

⚡ New to this?

This news is about a tool that watches what a Python app does while it runs. MCP, or Model Context Protocol, is a way for AI apps and agents to talk to tools and data sources, so monitoring it can show which tools were called, when sessions opened or closed, and what modules were loaded.

That matters because runtime behavior often reveals risks that source code alone does not. For a non-expert, the main takeaway is that BlueRock is trying to make AI server activity easier to audit after deployment, not just before release.

🦞 OpenClaw angle

If you run Python-based MCP agents, add a runtime sensor like this to your test and staging builds so you can see every tool call and import before production. Use the import hash data to flag unexpected dependency changes between releases, especially for agent tooling that pulls in many transitive packages. If you already forward logs to a SIEM, route the NDJSON output there so you can correlate MCP activity with auth, network, and process events.

BlueRock is aimed at security teams, AppSec engineers, and AI developers deploying MCP agents. The company says it is designed to answer a basic question: what is the application actually doing at runtime?

For MCP coverage, BlueRock records six event types. Those include server startup, tool or resource registration, client connections, session creation and termination, and protocol events such as requests, responses, and notifications.

The project says those events cover transports including stdio, HTTP, and SSE. It also says the monitoring is built into the OSS release, while policy enforcement and remediation are reserved for the full version.

BlueRock also tracks Python imports. According to the project, each module import is logged with the module name, file path, installed version when available, and a SHA-256 hash of the module on disk. The company says that gives users a way to verify imports and detect tampering between runs.

The sensor writes structured JSON events in NDJSON format to files under ~/.bluerock/event-spool/. The project says those logs can be inspected with jq, sent to Grafana and Loki, or forwarded into Datadog, Splunk, or a SIEM through OTLP.

The company says the OSS setup is meant to be simple. Users install the packages, create a small config file under ~/.bluerock/, and run scripts with python -m bluepython --oss. BlueRock says this works without adding imports, SDK calls, or application-side logging.

BlueRock also publishes a dashboard stack for local review. According to the project, the Grafana dashboard shows event timelines, tool call breakdowns, and session lifecycle panels.

The release notes say the open source version supports Python 3.10 and newer, with prebuilt wheels for Linux and macOS on both Intel and Apple Silicon. For platforms without wheels, the project says users can build from source with a Rust toolchain.

BlueRock is open source under Apache 2.0. The project says the sensor and its hooks can be inspected and extended, and that additional hook categories are available in the full version.

Source: HN Show HN ↗

More from Security News