Skip to content

v0.2.0

Choose a tag to compare

@maindala-it maindala-it released this 18 Aug 01:32
· 7 commits to main since this release

maindala 0.2.0

  • maindala scan [path] — discovers AI agents and MCP integrations in a local codebase.
    No account or API key required for the local report, and it completes even offline.
    Three confidence tiers: dependency manifests (package.json/requirements.txt/
    pyproject.toml/go.mod, high confidence), MCP client config / crew-agent definition
    files (.mcp.json/claude_desktop_config.json/crew.yaml/agents.yaml, high
    confidence), and source-code patterns (StateGraph(, Crew(, new Client(, low
    confidence — never auto-registered, even with --register). MCP servers found in a
    config file are matched against the public catalog for context (trust status, check
    count) via one unauthenticated lookup per server.
  • --format sarif --output <path> emits an OASIS SARIF 2.1.0 report plus a sidecar
    manifest.json (SHA-256 digest, versioned format) — a local, signable record the caller
    can hand to their own signing infrastructure (cosign, GPG, an in-house TSA) before
    anything is pushed to mAIndala. Output is byte-deterministic across repeated runs against
    unchanged input (sorted results, no wall-clock timestamp in the signed body,
    repo-relative paths only — see the test suite's dedicated determinism checks).
  • --timestamp additionally fetches a free RFC 3161 timestamp token over the manifest
    digest from a public Time-Stamp Authority (https://freetsa.org/tsr by default,
    overridable via MAINDALA_TSA_URL), anchoring the local record to a third party without
    mAIndala retaining anything.
  • --register --org <slug> pushes findings into that org's governed agent registry
    (requires maindala login <mk_...> as a member of the org — not an admin-only action).
    Metadata only by default; --include-definitions additionally sends the specific
    evidence that triggered each finding, enabling server-side trust scanning.
  • DATA.md — documents exactly what every command transmits, and to whom, including the
    new per-flag breakdown for scan.