Security fixes are backported to the current stable minor version only.

Do not open a public GitHub issue for security vulnerabilities.

To report a vulnerability privately:

1. Email arsiemuha@gmail.com with the subject line [BridgeMCP Security] <brief description>.
2. Include:
   • A description of the vulnerability
   • Steps to reproduce it
   • The version of BridgeMCP affected
   • Any suggested mitigations you are aware of

You will receive an acknowledgment within 72 hours. We aim to publish a fix within 14 days of confirmation, coordinated with the reporter.

If you prefer encrypted communication, request a PGP key in your initial email.

Once a fix is released, a security advisory will be published on the GitHub repository. Credit will be given to the reporter unless they prefer to remain anonymous.

We follow responsible disclosure: we will not publish details of a vulnerability until a fix is available, and we will notify downstream plugin authors if the vulnerability affects the plugin API.

The following are considered in-scope for security reports:

• Vulnerabilities in the BridgeMCP Python package (bridgemcp)
• Vulnerabilities in the MCP adapter that could allow a malicious MCP client to affect the server process
• Dependency vulnerabilities in pydantic or mcp that affect BridgeMCP users

The following are out of scope:

• Vulnerabilities in user-written tool, resource, or prompt handlers
• Vulnerabilities in third-party plugins
• Vulnerabilities in AI clients (Claude Desktop, Cursor, etc.)
• Authentication weaknesses in the HTTP transport — BridgeMCP does not provide authentication; that is the responsibility of a plugin or a reverse proxy

HTTP transport: app.run_http() binds to 127.0.0.1 by default and does not implement authentication. Do not expose it on 0.0.0.0 without a reverse proxy and authentication layer in front of it.

Tool handler inputs: BridgeMCP forwards arguments from MCP clients directly to your tool handlers. Validate all inputs inside your handlers if the tool is callable by untrusted clients.

Dependency pinning: BridgeMCP pins pydantic>=2.0 and mcp>=1.0. Pin your own dependencies with exact versions in production deployments to avoid unexpected updates.