wazuh-mcp changelog

Full Changelog: https://github.com/lidless-labs/wazuh-mcp/compare/v1.1.3...v1.1.4

Full Changelog: https://github.com/lidless-labs/wazuh-mcp/compare/v1.1.2...v1.1.3

Full Changelog: https://github.com/lidless-labs/wazuh-mcp/compare/v1.1.1...v1.1.2

Patch release for wazuhctrl.

• Adds repeatable npm publishing with audit and package payload checks.
• Keeps wazuhctrl and compatibility entrypoints intact.
• Updates release metadata after the ctrl conversion branch.

Security hardening release. The headline change: TLS certificate verification is now ON by default for both the Wazuh manager and the Wazuh Indexer connections. The previously published 1.0.0 shipped with verification off by default; if you rely on a self-signed lab certificate, you must now opt out explicitly with WAZUH_VERIFY_SSL=false and/or WAZUH_INDEXER_VERIFY_SSL=false (the server prints a startup warning when you do).

Security

• Verify TLS certificates by default for the manager and indexer clients; disabling verification is now an explicit opt-out that logs a startup warning (security: verify TLS by default).
• Gate unredacted get_manager_config output behind the server-side WAZUH_ALLOW_SENSITIVE_CONFIG flag; a model-supplied tool argument can never enable it on its own.
• Delimit attacker-influenced SIEM content returned to the model: alert full_log, alert rule_description, raw event data, and manager log descriptions are wrapped in <untrusted_siem_data> markers with an output.untrusted_data_note warning, and the affected tool descriptions flag the fields as data, never instructions.
• Route all tool-level error returns through the safe-error sanitizer so errors that bypass the client wrappers (JSON parse errors with body snippets, URL errors) never reach the MCP client raw.
• Fail fast at startup when WAZUH_INDEXER_URL is set without WAZUH_INDEXER_PASSWORD instead of silently sending an empty password.
• Minimize sensitive tool output by default: agent IPs, alert full logs, raw event data, process command lines, file hashes, and manager log descriptions are hidden unless opted in per call.
• Validate all MCP tool inputs with strict schemas: bounded pagination, length-limited search text, per-tool sort enums, and allowlisted identifiers for agent, alert, group, and SCA policy IDs.
• Encode Wazuh API path segments to prevent path injection.
• General security hardening pass across clients and tools, including sanitized diagnostics output that redacts URLs and never returns credentials.

Added

• Indexer-backed vulnerability tools: list_vulnerabilities and search_vulnerabilities.
• Response size caps via WAZUH_MCP_MAX_RESPONSE_BYTES (default 250000); oversized responses return a truncated JSON preview with output.response_truncated metadata instead of an error.
• pagination object (total, limit, offset, has_more) on paginated tool responses, alongside the existing top-level fields.
• Transient-error retries for manager GET and indexer search requests on 429, 502, 503, 504, and common network reset or timeout errors.
• AGENTS.md contributor guide and a scripts/verify entrypoint that runs test, typecheck, and build in order.

Changed

• The MCP server version reported in handshakes is now derived from package.json instead of a hardcoded constant.
• Documentation and test fixtures use RFC 5737 documentation addresses (192.0.2.x) instead of RFC 1918 space.
• Dependencies refreshed; npm audit clean.

Fixed

• Strip the draft-07 $schema marker the MCP SDK stamps on tool schemas, which some clients reject when listing the full tool set.

CI

• Publish with npm provenance (npm publish --provenance with id-token: write).
• Skip npm publish when the version already exists on the registry, making tag builds idempotent.

Full Changelog: https://github.com/solomonneas/wazuh-mcp/commits/v1.0.0