CI/CD Integration

The FusionTester runs entirely in RAM with zero external dependencies. No LLM API keys. No servers. No network. This makes it the first AI governance tool that fits natively into any CI/CD pipeline.

GitHub Actions

Basic

# .github/workflows/mva-audit.yml
name: MVA Governance Audit

on:
  push:
    branches: [main]
  pull_request:
    branches: [main]

jobs:
  audit:
    runs-on: ubuntu-latest
    steps:
      - uses: actions/checkout@v4
      - uses: actions/setup-node@v4
        with:
          node-version: 20
          cache: 'npm'

      - run: npm ci
      - run: npx vitest run --reporter=verbose

With Coverage + Artifacts

name: MVA Governance Audit

on: [push, pull_request]

jobs:
  audit:
    runs-on: ubuntu-latest
    steps:
      - uses: actions/checkout@v4
      - uses: actions/setup-node@v4
        with:
          node-version: 20
          cache: 'npm'

      - run: npm ci

      # Run all governance tests with coverage
      - run: npx vitest run --reporter=verbose --coverage

      # Upload coverage as artifact
      - uses: actions/upload-artifact@v4
        if: always()
        with:
          name: coverage-report
          path: coverage/

Separate Jobs per Governance Concern

name: MVA Governance Audit

on: [push, pull_request]

jobs:
  egress-firewall:
    name: "SOC2 CC6.1 — Egress Firewall"
    runs-on: ubuntu-latest
    steps:
      - uses: actions/checkout@v4
      - uses: actions/setup-node@v4
        with: { node-version: 20, cache: 'npm' }
      - run: npm ci
      - run: npx vitest run tests/firewall/ --reporter=verbose

  access-control:
    name: "SOC2 CC6.3 — Access Control"
    runs-on: ubuntu-latest
    steps:
      - uses: actions/checkout@v4
      - uses: actions/setup-node@v4
        with: { node-version: 20, cache: 'npm' }
      - run: npm ci
      - run: npx vitest run tests/guards/ --reporter=verbose

  system-rules:
    name: "Context Governance"
    runs-on: ubuntu-latest
    steps:
      - uses: actions/checkout@v4
      - uses: actions/setup-node@v4
        with: { node-version: 20, cache: 'npm' }
      - run: npm ci
      - run: npx vitest run tests/rules/ --reporter=verbose

  ui-blocks:
    name: "Response Quality"
    runs-on: ubuntu-latest
    steps:
      - uses: actions/checkout@v4
      - uses: actions/setup-node@v4
        with: { node-version: 20, cache: 'npm' }
      - run: npm ci
      - run: npx vitest run tests/blocks/ --reporter=verbose

This gives you separate check marks on every PR:

✅ SOC2 CC6.1 — Egress Firewall (3 tests, 0.4s)
✅ SOC2 CC6.3 — Access Control (4 tests, 0.5s)
✅ Context Governance (3 tests, 0.3s)
✅ Response Quality (3 tests, 0.3s)

With JUnit Report (for GitHub PR annotations)

- run: npx vitest run --reporter=junit --outputFile=results.xml
- uses: mikepenz/action-junit-report@v4
  if: always()
  with:
    report_paths: results.xml
    check_name: "MVA Governance Results"

GitLab CI

# .gitlab-ci.yml
stages:
  - test

mva-audit:
  stage: test
  image: node:20
  script:
    - npm ci
    - npx vitest run --reporter=verbose --reporter=junit --outputFile=results.xml
  artifacts:
    when: always
    reports:
      junit: results.xml

Azure DevOps Pipelines

# azure-pipelines.yml
trigger:
  - main

pool:
  vmImage: 'ubuntu-latest'

steps:
  - task: NodeTool@0
    inputs:
      versionSpec: '20.x'

  - script: npm ci
    displayName: 'Install dependencies'

  - script: npx vitest run --reporter=verbose --reporter=junit --outputFile=results.xml
    displayName: 'Run MVA Governance Audit'

  - task: PublishTestResults@2
    inputs:
      testResultsFormat: 'JUnit'
      testResultsFiles: 'results.xml'
      testRunTitle: 'MVA Governance Audit'
    condition: always()

Pre-Commit Hook

Block commits that break governance:

# .husky/pre-commit
npx vitest run tests/firewall/ --bail 1

Or with lint-staged:

// package.json
{
    "lint-staged": {
        "src/views/**/*.ts": "npx vitest run tests/firewall/ --bail 1",
        "src/agents/**/*.ts": "npx vitest run tests/guards/ --bail 1"
    }
}

Pull Request Status Checks

Required Checks

Configure your repository to require the governance audit to pass before merging:

1. Go to Settings → Branches → Branch protection rules
2. Enable "Require status checks to pass before merging"
3. Add these checks:
   • SOC2 CC6.1 — Egress Firewall
   • SOC2 CC6.3 — Access Control

Now no PR can merge if PII leaks or auth gates are broken.

Badge

Add a governance badge to your README:

[![MVA Audit](https://github.com/your-org/your-repo/actions/workflows/mva-audit.yml/badge.svg)](https://github.com/your-org/your-repo/actions/workflows/mva-audit.yml)

Performance in CI

The FusionTester is designed for CI speed:

Metric	Value
Cold start (npm ci + vitest)	~3 seconds
Per-test execution	~2ms
50 tests total	~100ms
Memory usage	~50 MB
API tokens consumed	0
API keys required	None
External services	None

Zero Flakiness

Because the FusionTester runs in RAM with zero external dependencies, your CI tests have zero flakiness. No API rate limits. No network timeouts. No model behavior variance. The same test produces the same result every time, on every machine.

---

SOC2 Automated Audit — The Tests That Block PRs

This is where MCP Fusion does what no other AI framework can. Each test below maps directly to a SOC2 control. If any of them fail, the PR cannot merge.

CC6.1 — Logical Access: No PII Reaches the LLM

// tests/firewall/user.firewall.test.ts
describe('SOC2 CC6.1 — PII Egress Prevention', () => {

    it('passwordHash NEVER reaches the LLM', async () => {
        const result = await tester.callAction('db_user', 'find_many', { take: 10 });

        for (const user of result.data as any[]) {
            // This is not a hope. This is a mathematical proof.
            // The field is physically absent — not masked, not hidden, REMOVED.
            expect(user).not.toHaveProperty('passwordHash');
        }
    });

    it('tenantId NEVER leaks across tenant boundaries', async () => {
        const result = await tester.callAction('db_user', 'find_many', { take: 10 });

        for (const user of result.data as any[]) {
            expect(user).not.toHaveProperty('tenantId');
        }
    });

    it('SSN NEVER appears in API responses', async () => {
        const result = await tester.callAction('db_employee', 'find_many', { take: 5 });

        for (const employee of result.data as any[]) {
            expect(employee).not.toHaveProperty('ssn');
            expect(employee).not.toHaveProperty('socialSecurityNumber');
        }
    });

    it('MCP transport contains ZERO sensitive fields', async () => {
        const result = await tester.callAction('db_user', 'find_many', { take: 5 });

        // The WIRE payload — what actually leaves the server
        const wirePayload = JSON.stringify(result.rawResponse);
        expect(wirePayload).not.toContain('passwordHash');
        expect(wirePayload).not.toContain('bcrypt');
        expect(wirePayload).not.toContain('tenantId');
        expect(wirePayload).not.toContain('ssn');
    });
});

CC6.3 — Access Control: Unauthorized Users Are Blocked

// tests/guards/user.guard.test.ts
describe('SOC2 CC6.3 — Role-Based Access Control', () => {

    it('GUEST cannot read user data', async () => {
        const result = await tester.callAction(
            'db_user', 'find_many', { take: 5 },
            { role: 'GUEST' },
        );
        expect(result.isError).toBe(true);
        expect(result.data).toContain('Unauthorized');
    });

    it('GUEST cannot create users', async () => {
        const result = await tester.callAction(
            'db_user', 'create',
            { email: 'hack@evil.com', name: 'Hacker' },
            { role: 'GUEST' },
        );
        expect(result.isError).toBe(true);
    });

    it('GUEST cannot delete users', async () => {
        const result = await tester.callAction(
            'db_user', 'delete', { id: '1' },
            { role: 'GUEST' },
        );
        expect(result.isError).toBe(true);
    });

    it('ADMIN CAN read user data', async () => {
        const result = await tester.callAction(
            'db_user', 'find_many', { take: 5 },
            { role: 'ADMIN' },
        );
        expect(result.isError).toBe(false);
        expect((result.data as any[]).length).toBeGreaterThan(0);
    });

    it('every action is protected — no gaps', async () => {
        const actions = ['find_many', 'create', 'update', 'delete'];
        for (const action of actions) {
            const result = await tester.callAction(
                'db_user', action, {},
                { role: 'GUEST' },
            );
            expect(result.isError).toBe(true);
        }
    });
});

CC6.7 — Output Controls: Only Declared Fields Exist

// tests/firewall/order.firewall.test.ts
describe('SOC2 CC6.7 — Output Field Control', () => {

    it('internal fields are stripped from order responses', async () => {
        const result = await tester.callAction('db_order', 'find_many', { take: 5 });

        for (const order of result.data as any[]) {
            // Business logic fields — ALLOWED
            expect(order).toHaveProperty('id');
            expect(order).toHaveProperty('total');
            expect(order).toHaveProperty('status');

            // Internal fields — STRIPPED
            expect(order).not.toHaveProperty('internalNotes');
            expect(order).not.toHaveProperty('profitMargin');
            expect(order).not.toHaveProperty('costPrice');
            expect(order).not.toHaveProperty('supplierEmail');
        }
    });
});

CC7.1 — System Operations: Governance Directives Present

// tests/rules/user.rules.test.ts
describe('SOC2 CC7.1 — LLM Governance Directives', () => {

    it('PII handling rules are injected on user responses', async () => {
        const result = await tester.callAction('db_user', 'find_many', { take: 1 });

        expect(result.systemRules).toContain(
            'Email addresses are PII. Mask when possible.'
        );
    });

    it('financial precision rules are injected on order responses', async () => {
        const result = await tester.callAction('db_order', 'find_many', { take: 1 });

        expect(result.systemRules).toContain(
            'CRITICAL: amount_cents is in CENTS. Divide by 100 for display.'
        );
    });

    it('user rules do NOT appear in order responses (Context Tree-Shaking)', async () => {
        const result = await tester.callAction('db_order', 'find_many', { take: 1 });

        // Proves the LLM only sees relevant governance
        expect(result.systemRules).not.toContain('Email addresses are PII.');
    });
});

CC8.1 — Change Management: Resource Protection

// tests/guards/user.oom.test.ts
describe('SOC2 CC8.1 — Input Boundary Protection', () => {

    it('rejects take > 50 (prevents memory exhaustion)', async () => {
        const result = await tester.callAction('db_user', 'find_many', { take: 99999 });
        expect(result.isError).toBe(true);
    });

    it('rejects negative take (prevents negative indexing)', async () => {
        const result = await tester.callAction('db_user', 'find_many', { take: -1 });
        expect(result.isError).toBe(true);
    });

    it('rejects SQL injection in string fields', async () => {
        const result = await tester.callAction('db_user', 'create', {
            email: "admin@test.com'; DROP TABLE users; --",
            name: 'Hacker',
        });
        expect(result.isError).toBe(true); // Zod email validation rejects
    });
});

---

GitHub Actions Output

When these tests run in your pipeline, the PR shows:

✅ SOC2 CC6.1 — PII Egress Prevention          4 tests passed (0.3s)
✅ SOC2 CC6.3 — Role-Based Access Control       5 tests passed (0.4s)
✅ SOC2 CC6.7 — Output Field Control            1 test passed  (0.2s)
✅ SOC2 CC7.1 — LLM Governance Directives       3 tests passed (0.2s)
✅ SOC2 CC8.1 — Input Boundary Protection       3 tests passed (0.2s)
──────────────────────────────────────────────────────────────────
   Total: 16 tests | 0 failures | 1.3s | $0.00 in API tokens

If a developer introduces a regression that leaks passwordHash:

✅ SOC2 CC6.3 — Role-Based Access Control       5 tests passed (0.4s)
✅ SOC2 CC6.7 — Output Field Control            1 test passed  (0.2s)
✅ SOC2 CC7.1 — LLM Governance Directives       3 tests passed (0.2s)
✅ SOC2 CC8.1 — Input Boundary Protection       3 tests passed (0.2s)
❌ SOC2 CC6.1 — PII Egress Prevention          1 FAILED (0.3s)
   FAIL: passwordHash NEVER reaches the LLM
         Expected: not toHaveProperty('passwordHash')
         Received: { id: '1', name: 'Alice', passwordHash: 'bcrypt$abc' }
──────────────────────────────────────────────────────────────────
   ❌ PR BLOCKED — PII leak detected. Cannot merge.

The PR cannot merge. The passwordHash leak is caught before it reaches production, before it reaches the LLM, and before your compliance officer sees it.

This is What Automated SOC2 Looks Like

No other AI framework in existence can do this. Every other MCP server relies on a developer opening Claude Desktop and saying "looks fine to me." That is not an audit. That is a liability.

MCP Fusion converts compliance requirements into expect() assertions that run on every commit. The auditor receives a CI/CD log, not a verbal assurance.

Manual Audit vs Automated Audit

	Manual SOC2 Audit	MCP Fusion Automated Audit
Frequency	Quarterly or annually	Every commit, every PR
Cost	$50,000–$250,000 per audit	$0.00 (CI/CD minutes only)
Evidence	Screenshots, spreadsheets, interviews	Deterministic test results, CI logs
Time to detect leak	3–12 months	< 60 seconds
Coverage	Sample-based (5–10% of code)	100% of every MVA pipeline
Reproducibility	Non-reproducible (point-in-time)	Reproducible on every machine
Proof format	PDF report with checkboxes	expect().not.toHaveProperty()