MCP Server

Verification your AI agents can call — and your compliance team can trust.

nexus2AI ships as an MCP server: a single secure endpoint that turns document verification into a native tool for any AI agent, while enforcing human oversight and on-premise data handling at the protocol level.

Connect via MCP Book a Compliance Walkthrough

What your agent gets

Four tools, each with a built-in compliance guarantee.

Tool	What it does	Compliance note
`verify_document`	Submits a document for verification.	Returns a job ID — nothing is auto-finalised.
`get_verification_result`	Returns accepted / rejected / needs_review with per-field confidence.	needs_review cannot be coerced to approved.
`get_consensus_log`	Returns the full multi-model debate.	Gives auditors the complete reasoning.
`list_ontologies`	Lists the document types the agent may verify.	Scopes exactly what an agent can check.

verify_document

Submits a document for verification.

Returns a job ID — nothing is auto-finalised.

get_verification_result

Returns accepted / rejected / needs_review with per-field confidence.

needs_review cannot be coerced to approved.

get_consensus_log

Returns the full multi-model debate.

Gives auditors the complete reasoning.

list_ontologies

Lists the document types the agent may verify.

Scopes exactly what an agent can check.

Autonomy with a hard stop

Most “AI verification” lets the model decide. nexus2AI does not. The needs_review verdict is terminal in the tool contract — an agent receives it, surfaces it, and stops. Only a human can resolve it. Combined with on-premise processing and a council that never sees raw PII, this is verification you can put in front of an auditor and an autonomous agent at the same time.

Lane

Agent

Calls verify_document and polls for the result.
Acts automatically on accepted and rejected.
On needs_review (confidence below threshold) the agent stops.

Hard stop

Lane

Human reviewer

Uncertain documents land in the review queue.
A person inspects field-level evidence and the consensus log.
Only the human can resolve needs_review.

Set up in three steps

Generate an API key

Create a scoped key from your nexus2AI workspace.

Point your agent at the MCP endpoint

Add the single secure nexus2AI MCP endpoint to your agent's config.

Tools appear automatically

The verification tools register themselves — no custom integration build.

ClaudeChatGPTMicrosoft CopilotAny MCP-compatible agent framework

If it speaks MCP, it can verify with nexus2AI.

Not using agents yet? The same verification is available through our REST API — see the integration guide.