Skip to content

Security: BlockRunAI/blockrun-mcp

Security

SECURITY.md

Security Policy

BlockRun MCP signs real-money transactions (x402 USDC payments on Base, pUSD trading on Polygon) with a locally held key. We treat every report that could lead to loss of funds, key exposure, or silent mis-reporting of a money operation as the highest priority.

Reporting a vulnerability

Please do not open a public issue for security reports.

Use GitHub's private vulnerability reporting: go to the repository's Security tab → Report a vulnerability (or https://github.com/BlockRunAI/blockrun-mcp/security/advisories/new). Reports go directly and privately to the maintainers.

Include what you can of: the affected tool/action (e.g. blockrun_polymarket action:"withdraw"), the version (npm ls @blockrun/mcp or the version field in package.json), reproduction steps, and impact — especially whether funds can be moved, stranded, or mis-reported.

What to expect

  • Acknowledgement within 72 hours.
  • An assessment and remediation plan within 7 days for anything fund-affecting.
  • Fixes ship as a patch release to npm and the MCP registry (this repo's publish pipeline releases both from main); running npx installations pick the fix up on their next cold start.
  • Credit in the release notes and CHANGELOG unless you ask otherwise.

Supported versions

Only the latest release on npm (@blockrun/mcp@latest) is supported. The recommended install (npx -y @blockrun/mcp@latest) resolves to it on every cold start, so there is no supported back-catalog to patch.

Scope notes

  • The wallet key lives in ~/.blockrun/.session (or BLOCKRUN_WALLET_KEY) and must never leave the machine; any code path that would print, log, or transmit it is in scope and critical.
  • Polymarket trading approvals, redeem, and withdraw paths move user funds — incorrect success reporting ("the transaction landed" presented as "you were paid") counts as a vulnerability here, not just a bug.
  • The hosted gateway (blockrun.ai) and its endpoints are a separate codebase; reports about them are still welcome through the same channel and will be routed to the right place.

There aren't any published security advisories