MCP Server

Three tool calls. Every MCP host.

Witniumchain ships a native Model Context Protocol server. Any agent framework, AI assistant, or platform that speaks MCP can witness, verify, and bind outputs to a tamper-evident chain — with zero glue code.

Read the MCP docs Get credentials

The three tools

witness · verify · bind

witness

Create a tamper-evident record of an AI output, signed and sealed on-chain.

await mcp.call('witness', {
  dataId: sha256(output),
  context: { model, actor }
});

verify

Check a witness against the chain. Returns the full signature chain and on-chain proof.

await mcp.call('verify', {
  witnessId: 'wc_...'
});

bind

Bind multiple witnesses or attestations into an evidence chain. For multi-party workflows.

await mcp.call('bind', {
  witnessIds: [...],
  signers: [...]
});

Tool surface

How the LLM sees it.

Tool descriptions tuned for natural LLM tool-use. Your agent picks up Witniumchain the same way it picks up any other MCP capability — no prompt engineering required.

# Tool: witness
Records a cryptographic witness for any data identifier.
Use when the user asks for a tamper-evident record, audit
log, or compliance receipt of an AI output.

Inputs:
  dataId: SHA-256 hash of the artifact (string, required)
  context: model, actor, source metadata (object)
  signers: optional list of co-signers (array)

Output:
  witnessId: chain-anchored identifier
  proof: verification URL

Hosts

Drops into every MCP client.

Claude Desktop

Anthropic's reference client. Add the server block to your config.

Cursor

IDE-native AI. Witniumchain shows up alongside your other tools.

Any MCP host

OAuth 2.1 + PKCE. Spec-compliant — if it speaks MCP, it works.