darktrace-mcp

https://github.com/nuoframework/darktrace-mcp is public.

Git history records who changed what and when; all changes land through reviewed pull requests.

Work-in-progress branches and pull requests are pushed to the public repository before release (for example https://github.com/nuoframework/darktrace-mcp/pulls?q=is%3Apr).

Git.

Every release has a unique Semantic Versioning identifier (1.0.0, 1.1.0, 1.1.1) and a git tag vX.Y.Z: https://github.com/nuoframework/darktrace-mcp/tags

Semantic Versioning 2.0.0; package.json version equals the tag, enforced by the release workflow.

Tags vX.Y.Z: https://github.com/nuoframework/darktrace-mcp/tags

CHANGELOG.md (https://github.com/nuoframework/darktrace-mcp/blob/main/CHANGELOG.md) and GitHub Releases (https://github.com/nuoframework/darktrace-mcp/releases), each release listing changes and known limitations.

No publicly known vulnerabilities have been fixed yet. CHANGELOG.md and the GitHub Release notes record security-relevant changes and will identify any fixed vulnerability (CVE where available).

Bugs are reported through GitHub Issues (https://github.com/nuoframework/darktrace-mcp/issues); vulnerabilities through the private process in SECURITY.md.

GitHub Issues: https://github.com/nuoframework/darktrace-mcp/issues

The maintainer acknowledges bug reports; the project is new (first public release 2026-10-06) and all reports received so far have been handled.

Enhancement requests are triaged in GitHub Issues; the project aims to respond within 14 days.

GitHub Issues and pull requests are a public, searchable archive: https://github.com/nuoframework/darktrace-mcp/issues?q=is%3Aissue

SECURITY.md (https://github.com/nuoframework/darktrace-mcp/blob/main/SECURITY.md): GitHub private vulnerability reporting via the Security tab / https://github.com/nuoframework/darktrace-mcp/security/advisories/new

GitHub private vulnerability reporting is enabled for the repository: https://github.com/nuoframework/darktrace-mcp/security/advisories/new (see https://github.com/nuoframework/darktrace-mcp/blob/main/SECURITY.md). Reports are encrypted in transit over TLS and visible only to the maintainer and the reporter; SECURITY.md asks reporters not to use public issues.

SECURITY.md commits to an initial response to vulnerability reports within 14 days.

npm run build (scripts/build.mjs, TypeScript compilation and catalogue generation), documented in CONTRIBUTING.md and docs/getting-started.md.

Node.js 22+, npm and the TypeScript compiler; no custom build system.

Node.js, npm, TypeScript and ESLint are all FLOSS.

npm test runs 230 functional tests (unit, contract, MCP protocol, CLI) with Node's built-in test runner; npm run test:security runs a 1,150-case isolated security suite (test/security).

npm test and npm run test:security (documented in CONTRIBUTING.md); both run in CI.

The functional suites cover the policy, approval, client, config, CLI and server layers; the security suite exercises every API operation and denial path (1,150 subcases). Code coverage is not yet measured.

https://github.com/nuoframework/darktrace-mcp/blob/main/.github/workflows/ci.yml runs lint, typecheck, functional and security suites on Node 22 and 24 plus native Docker gates on amd64 and arm64 for every push and pull request; branch protection requires them.

CONTRIBUTING.md requires tests with every change; CI pins the expected test counts so missing tests are detected.

Recent changes added tests with the functionality (the functional suite grew from 130 to 230 tests and the security suite from 325 to 1,150 cases during the 1.1.x releases).

https://github.com/nuoframework/darktrace-mcp/blob/main/CONTRIBUTING.md ("How changes are accepted").

TypeScript strict mode (tsconfig.json) and ESLint with @eslint/js, typescript-eslint and eslint-plugin-security (eslint.config.mjs) run in CI (lint job).

The lint job fails on any error or warning; the tree is clean.

TypeScript strict; ESLint recommended + 13 security rules as errors.

The design follows least privilege (read-only by default, operator profiles), fail-closed policy checks, human approval for critical actions, defence in depth and input/output bounding: docs/security/threat-model.md, threat-model-writes.md and design-decisions.md.

The maintainers reviewed OWASP Top 10 for LLM applications, the OWASP MCP Top 10 and MCP-specific attack research (prompt injection via tool results, tool poisoning, confused deputy, SSRF, TLS downgrade): docs/security/mcp-attack-research.md and mcp-attack-research-round2.md.

TLS from Node.js/OpenSSL and HMAC-SHA1 request signing as required by the Darktrace API, both via node:crypto / node:tls; no custom algorithms.

Only node:crypto (createHmac, randomUUID, createHash) and node:tls/https are used; no cryptography is implemented in the project.

OpenSSL through Node.js, both FLOSS.

TLS uses Node.js defaults (TLS 1.2+/1.3, ≥128-bit ciphers); the server rejects any flag or environment variable that would weaken them. HMAC keys are the appliance-issued API tokens.

TLS 1.2+/1.3 only; SHA-256 for all project checksums and digests. SHA-1 appears only inside HMAC-SHA1, which the Darktrace API mandates for request signing and for which no practical attack exists.

No MD5, RC4, DES or SHA-1 signatures; HMAC-SHA1 is used solely because the vendor API requires it (HMAC-SHA1 is not affected by SHA-1 collision attacks). Checksums and audit chains use SHA-256.

TLS connections use Node.js default cipher suites (ECDHE, TLS 1.3), which provide forward secrecy; cipher overrides are refused at startup.

The project stores no passwords or credentials; API tokens stay in operator-owned files (mode 0600) and are never written by the project.

node:crypto randomUUID() and randomBytes() for preview handles, request identifiers and approval state keys.

Delivered over HTTPS from npm (with SLSA provenance), ghcr.io (by immutable digest) and GitHub Releases with SHA256SUMS; the wizard pins absolute paths and digests.

npm packages carry Sigstore provenance attestations; releases publish SHA256SUMS and image digests, so a user can verify integrity before use.

No publicly known vulnerabilities are open; SECURITY.md commits to fixing confirmed vulnerabilities promptly.

No known critical vulnerabilities.

No credentials in the repository: tokens are read from operator files only, tests assert that token values never appear in logs or outputs, and GitGuardian secret scanning runs on pull requests.

CodeQL (security-extended) on every push/PR and weekly (.github/workflows/codeql.yml) and ESLint with eslint-plugin-security in the CI lint job.

CodeQL security-extended queries cover common JavaScript/TypeScript weaknesses (injection, path traversal, insecure randomness, etc.).

All CodeQL findings were fixed (release scripts) or dismissed with a documented reason (lab-only tooling); docs/security/supply-chain-checks.md.

On every push and pull request, plus a weekly schedule.

A 1,150-case security suite exercises the running server with hostile inputs (prompt-injection corpora, malformed frames, oversized bodies, fault injection, real TLS) before each release; see docs/security/adversarial-results-writes.md and mcp-attack-research-round2.md.

TypeScript/JavaScript on Node.js is memory-safe; no unsafe-language components.

Tests use node:assert/strict throughout, and production code validates every input with strict schemas (zod) and fails closed.

Findings from the adversarial campaigns (AD-W-01..21, AD2-01..04) were fixed or reconciled before release: docs/security/adversarial-results-writes.md.