4. Core Components

The Osirium MCP Node is composed of three functional layers that work together to transform AI-generated content into cryptographically signed attestations:

---

4.1 Prompt Inference Layer

This layer handles prompt ingestion and AI interaction.

• Prompts can be manually submitted via CLI or programmatically fed into the system.

• The node uses Claude AI via API to process each prompt and generate a natural language output.

• Prompt and response are both stored for attestation and auditability.

---

4.2 Attestation Layer

Once the AI output is received, the system:

• Structures the result as a JSON payload, including:

  • The original prompt

  • The AI-generated output

  • A timestamp

• Signs the payload using a local EVM private key, transforming it into a verifiable attestation.

• Outputs the result with:

  • prompt

  • output

  • signature

  • signer address

This enables third parties to verify the source and integrity of the output.

---

4.3 Node Configuration & Execution

Osirium nodes are fully configurable and modular.

• Each node is set up using a local config file (mcp.json) containing:

  • Ethereum private key

  • Claude API key

  • Node metadata (name, description, etc.)

• The main process (main.js) continuously handles:

  • Prompt fetching

  • AI inference

  • Attestation generation

Nodes can be deployed by any operator and used for either internal pipelines or as public services.

---