Skip to main content

Behavry Proxy (OSS)

Feature row 57 — Sprint OSS-Proxy (Partial)

Phase 1 ships today: the standalone proxy with the 9-step enforcement pipeline. Phase 2 (full test matrix + CI), Phase 3 (TUI polish), and Phase 4 (release engineering, packaging) are in progress.

What this is​

Behavry Proxy is the inline MCP governance proxy from the commercial Behavry platform, extracted into a standalone open-source project. It's Apache 2.0 licensed and published separately from this repo. Use it when you want the proxy enforcement layer — policy, DLP, behavioral monitoring, audit logging — without running the full Behavry control plane.

Typical uses:

  • A security-conscious team that wants per-tool-call policy enforcement on Claude Code / Cursor without signing up for SaaS
  • Embedding MCP governance in an existing internal platform
  • A starting point for contributing back features that later land in the commercial product

Architecture​

One binary, one config file. The proxy sits between an MCP client and any number of MCP servers:

MCP client (Claude Code / Cursor / ...) → behavry-proxy → MCP server(s)

On every tool call, the proxy runs a 9-step enforcement pipeline — the same pipeline the commercial product uses, stripped of the tenant / billing / enterprise features:

  1. Parse incoming MCP request
  2. Identify agent + session from headers or config
  3. Classify the tool via the built-in catalog
  4. Evaluate OPA policy (local OPA sidecar or embedded rego via a Go bridge)
  5. Scan for DLP patterns
  6. Scan for prompt injection
  7. Rate-limit if configured
  8. Forward or deny
  9. Audit to a local log

Each step is optional and configurable — you can run the proxy in lightweight mode (just log everything) or in hard-block mode.

Config​

# behavry-proxy.yaml
listen: 0.0.0.0:8080
servers:
  filesystem:
    command: ["npx", "-y", "@modelcontextprotocol/server-filesystem", "/workspace"]
  github:
    url: https://github-mcp.example/mcp

policy:
  mode: enforce        # report-only | enforce
  rego: ./policies/

dlp:
  enabled: true
  patterns_file: ./dlp.yaml

audit:
  sink: file
  path: ./audit.log

What's in Phase 1 (today)​

  • The 9-step pipeline
  • OPA REST client + bundled default policies
  • DLP scanner (port of the Python module)
  • Injection scanner (BIS port)
  • JSON audit log sink
  • Configurable rate limiting

What's Phase 2 / 3 / 4 (in progress)​

  • Full test matrix across MCP client types
  • Continuous integration publishing signed binaries
  • Interactive TUI for live policy tuning
  • Homebrew / apt / deb / rpm packaging
  • Windows support beyond WSL

Source and downloads​

  • Repo: github.com/behavry/behavry-proxy
  • Releases: signed binaries for macOS (arm64 + x86_64) and Linux (arm64 + x86_64)
  • License: Apache 2.0

Relationship to the commercial product​

Behavry Proxy is a subset of the commercial proxy. Everything in the OSS version also exists in the commercial product. Features that need a backend (multi-tenant, SIEM fan-out, compliance modules, dashboards) are not part of the OSS project — by design.

If you want the SaaS experience with dashboards, use the commercial product. If you want just the enforcement layer on your laptop or inside your own platform, use the OSS proxy.