Tobari MCP Server

The Tobari MCP (Model Context Protocol) Server is an interface that allows AI agents to directly read and verify digital credentials (Tobari documents), and create privacy-preserving Verifiable Presentations.

By introducing this server, AI agents gain the autonomous capability to perform high-level tasks, such as "Extracting only the name and address from a user's residence certificate to generate signed submission data."

Why AI needs Tobari?

Large Language Models (LLMs) are excellent at processing text, but they cannot inherently "trust" or "verify" the authenticity of data they receive from a user.

• Authenticity: Agents can verify that a document was actually signed by a specific authority.
• Privacy: Agents can minimize data leakage by only sharing specific fields required for a task (Selective Disclosure).
• Structure: Documents are machine-readable (CBOR/JSON), removing the ambiguity of free-text parsing.

Features

• Document Inspection: Allow AI to view the structure and content of a signed Tobari document.
• Signature Verification: Tools for the AI to check if the document integrity is maintained.
• Presentation Creation: Instruct the AI to generate a sub-credential (Verifiable Presentation) containing only necessary fields.

Tools Reference

The server exposes two categories of tools: Core Tools for handling credentials, and Demo Tools for development and simulation.

Core Tools (Business Logic)

Tool Name	Description
read_tobari_file	Reads and parses a Tobari file (.cose/.html). Verifies issuer signatures if a key is provided.
create_presentation	Creates a selective disclosure VP (Verifiable Presentation) from a document. Supports signerPath and fallbackToEphemeral.
preview_presentation	Decodes and summarizes a VP. Optional signature verification.
verify_presentation	Verifies a VP, checking both Issuer and Device (Holder) signatures. Supports PQC.
analyze_service_request	Analyzes a Service Request file to understand what credentials are required.
sign_with_jpki	Signs data using a physical My Number Card (macOS/Windows/Linux).
read_mynumber	Reads My Number from a physical card.
read_photo	Reads the face photo from a physical card.

Development & Demo Tools

These tools are prefixed with demo_ and are intended for testing, prototyping, and simulating ecosystem actors.

Tool Name	Description
demo_list_examples	Lists sample documents and keys available in the project's examples/ directory.

Note: Demo server tools were removed. Use preview_presentation for VP inspection and JSON previews.

Encrypted Documents

For encrypted Tobari files (including hybrid HPKE + PQC), pass a decrypt object to the relevant tools (read_tobari_file, create_presentation, prepare_presentation, analyze_service_request, demo_list_examples, generate_passport_zkp_input).

{
  "name": "read_tobari_file",
  "arguments": {
    "path": "/absolute/path/to/encrypted.cose",
    "decrypt": {
      "hpkeSecret": "tobari-demo-secret-key-32-bytes-long!!",
      "hpkeInfo": "tobari-storage-v1",
      "pqcPrivateKeyPath": "/absolute/path/to/recipient-pqc-private-key.json"
    }
  }
}

Defaults fall back to TOBARI_HPKE_SECRET, TOBARI_HPKE_INFO, and TOBARI_PQC_PRIVATE_KEY_PATH (or demo values if unset).

Getting Started

Installation

npm install -g @tobari/mcp-server

Configuration (Claude Desktop)

Add the following to your claude_desktop_config.json:

{
  "mcpServers": {
    "tobari": {
      "command": "tobari-mcp",
      "args": ["--key-dir", "/path/to/keys"]
    }
  }
}

Security

The MCP server never shares the user's private keys or full raw documents with the AI unless explicitly allowed. It operates based on the "Minimality Principle," ensuring that AI agents act only within the scope of delegated trust.

PQC Countersign Demo (verify_presentation)

This demo shows how to attach an ML-DSA-65 countersignature on the issuer side and verify it via verify_presentation.

1) Generate a PQC-signed credential

Manually run: bun examples/juminhyo/gen-tobari.ts --pqc

This creates:

• examples/juminhyo/juminhyo.cose
• examples/juminhyo/issuer-pqc-public-key.json

2) Create a presentation (VP)

Use create_presentation or your normal flow to generate a VP from juminhyo.cose.

Note: create_presentation supports signerPath (override signer binary path) and fallbackToEphemeral (use a temporary key if signer is unavailable).

3) Verify with PQC public key

Pass issuerPqcPublicKeys alongside classic issuer keys:

{
  "vpBase64": "<base64-vp>",
  "issuerPublicKeys": {
    "jp.v0.juminhyo": "/absolute/path/to/examples/juminhyo/issuer-key.json"
  },
  "issuerPqcPublicKeys": {
    "jp.v0.juminhyo": "/absolute/path/to/examples/juminhyo/issuer-pqc-public-key.json"
  }
}

The response includes issuerPqcPresent and issuerPqcValid for each document.

Quick VP Preview

Use preview_presentation to inspect the VP without running a demo server:

{
  "name": "preview_presentation",
  "arguments": {
    "vpBase64": "<base64-vp>"
  }
}

If you want to verify signatures at the same time, pass issuerPublicKeys (and optional issuerPqcPublicKeys).

Optional output controls:

{
  "name": "preview_presentation",
  "arguments": {
    "vpBase64": "<base64-vp>",
    "format": "readable",
    "redact": true,
    "maxStringLength": 120
  }
}

Set format to summary, readable, or full. Use includeDecoded: true to attach the decoded VP payload alongside the human-readable view.