Model Context Protocol

AI compliance inside the AI agents your developers already use

Claude Code, GitHub Copilot and GitHub Actions scan, fix and attest AI systems through one CertOne MCP server.

Recorded live on our production platform. Three AI agents take three AI systems from a failing scan to a signed attestation.

Three use cases, three AI agents

Technology use case

A RAG service, checked by Claude Code

A retrieval-augmented answering service about to be released. The developer types one sentence in the terminal and Claude does the rest.

Agent
Claude Code in the terminal, with Claude Opus 5.5
What CertOne found
Two hard-coded API keys in src/config.py, line 31
Laws and standards named
EU AI Act Art. 15GDPR Art. 32ISO/IEC 42001

Step by step

  1. ScanInside a hardware-attested confidential VM
  2. VerdictFAIL with 3 high findings, each with its law, file and line
  3. FixPatch from CertOne's sealed LLM agents, reviewed by a verifier
  4. ApplyClaude shows the diff and moves both keys to the environment
  5. Scan againPASS with no findings
  6. AttestRelease allowed and the signed attestation shown in the chat
Outcome

From FAIL to a signed attestation in one terminal conversation. The keys never reach the release, and the attestation is bound to that exact version.

Integrate with Claude Code
Healthcare use case

An ECG arrhythmia monitor, checked by GitHub Copilot in VS Code

An AI model that flags arrhythmias in ECG signals, regulated as a medical device. The developer asks Copilot in agent mode to check it before release.

Agent
GitHub Copilot in VS Code, agent mode, with Claude Opus 5.5
What CertOne found
A hard-coded model registry API key in azure-ml/publish_model.py, line 23
Laws and standards named
EU AI Act Art. 15GDPR Art. 32ISO/IEC 42001EU MDR Annex I, GSPR 17.2 and 17.4

Step by step

  1. ScanInside a hardware-attested confidential VM
  2. VerdictFAIL with 4 high findings, each with its law, file and line
  3. FixPatch from CertOne's sealed LLM agents, reviewed by a verifier
  4. ApplyCopilot applies the diff and the editor highlights the line
  5. Scan againPASS with no findings
  6. AttestRelease allowed and the signed attestation shown in the chat
Outcome

The credential is gone from the medical AI's code before release, and the evidence covers the medical device regulation as well as the AI rules.

Integrate with VS Code
Automotive use case

An ADAS driver monitor, certified in GitHub Actions

A driver monitoring model for an automotive ECU. A pull request starts the workflow, where an agent calls CertOne with no one at the keyboard.

Agent
Claude Code in a GitHub Actions workflow, on every pull request
What CertOne found
A live OTA backend API key in ota/publish_release.py, line 22
Laws and standards named
UNECE R155EU AI Act Art. 15GDPR Art. 32ISO/IEC 42001 Annex A.6

Step by step

  1. ScanInside a hardware-attested confidential VM
  2. VerdictFAIL with 4 high findings, each with its law, file and line
  3. FixPatch from CertOne's sealed LLM agents, reviewed by a verifier
  4. ApplyThe agent commits the fix to the pull request branch
  5. Scan againPASS with no findings
  6. AttestRelease allowed and the full report posted on the pull request
Outcome

Every pull request gets a verdict, a fix and a signed attestation before merge, recorded on the pull request itself.

Integrate with GitHub Actions

Four tools behind every step

certone_scan

Confidential scan of a project folder, returning the verdict and each finding with its law, file and line.

certone_fixes

The fixes proposed by CertOne's sealed LLM agents, with the verifier's review.

certone_release_gate

Allows a release only for a passing verdict with a signed attestation.

certone_attestation

The signed attestation of a passing report, bound to its exact version.

certone-mcp runs next to the agent with your tenant's API key. The server never approves or applies a fix on the platform. The developer reviews every change in their own tree, and only a new passing scan issues an attestation.

ChatGPT and your own agents

Any MCP client that runs local servers can use certone-mcp. The setup guide shows how with the OpenAI Agents SDK and the Claude Agent SDK.

Connect your own agent

See it on your own AI systems

Set it up in a few minutes, or talk to us about your own AI systems.

Read the setup guide Watch all the demos