Skip to content

Connect an agent

Add NIUA to your AI tool.

One line of setup. Then your AI checks its own files.

claude mcp add --transport http niua https://mcp.ohao.tech/mcp

The first connect opens a NIUA sign-in in your browser. To skip it, send an API key as a Bearer token.

What happens next

  1. 01

    Add the line

    Run the line above in your terminal. You do it once.

  2. List the NIUA contracts, then check ./chair.glb against glb.prop and fix whatever fails until it passes.
    02

    Ask it to check a file

    Say this, or anything like it, about your own file.

  3. 03

    It fixes what fails and gets a receipt

    NIUA says what failed and how to fix it. When the file passes, your AI gets a receipt: a signed check result anyone can verify.

Details for developers

NIUA is a hosted MCP server and a REST API. Add it once and your agent can check any artifact against a contract and read back what failed.

Checking your own file needs a client that can make HTTP requests, as Claude Code, Cursor and Codex can: the upload is an HTTP PUT of the file, not an MCP call. A chat app without one can still check NIUA's sample files.

Connect your coding agent to NIUA so it can check its own work and fix what fails. For developers who run Claude Code, Cursor, Codex or any MCP client.

  1. 01 Add NIUA to Claude Code

    claude mcp add --transport http niua https://mcp.ohao.tech/mcp
  2. 02 Ask it to check a file

    claude -p "Upload lamp.glb and check it against glb.prop@1 with NIUA. Fix what fails."
  3. 03 Anyone re-checks the receipt it gets ("valid": true)

    curl https://api.ohao.tech/api/v1/receipts/$RECEIPT_ID
MCP
the protocol agents such as Claude Code use to call tools. NIUA is an MCP server.
Gate
NIUA's checker. It runs a contract on your file and says pass or fail, with a fix for each fail.
Contract
the published rules a file must meet, by name and version. Each rule has a threshold and a fix.
Receipt
a signed note that this exact file passed this exact contract. Anyone can verify it, offline.

For agents: llms.txt · OpenAPI