OpenADA
Back to ADA guidance

OpenADA developer docs

Put accessibility checks inside your AI tools.

OpenADA exposes a public, stateless Model Context Protocol server. Connect it once, then ask ChatGPT, Codex, Claude Code, or another MCP client to check a page, queue a bounded website crawl, follow progress, and explore public scan history.

MCP endpointhttps://openada.us/mcp

Quick start

One endpoint, four tools

OpenADA returns structured results that AI clients can summarize and act on. Site scans are asynchronous, so a client receives a job ID immediately and polls durable progress instead of waiting on a long request.

Check a page

Run axe-core accessibility checks and LanguageTool-compatible language analysis against HTML, text, or a public URL.

Scan a website

Queue a same-host crawl with a configurable limit of up to 100 public pages.

Follow progress

Read pages scanned, pages discovered, current URL, errors, score, grade, and the completed report.

Browse the archive

Inspect public sites, scan history, page findings, and changes over time.

Connect a client

ChatGPT, Codex, and Claude

Use the instructions for the tool you already work in. The public demo is anonymous; protected deployments can require a bearer token or API key.

01

ChatGPT

ChatGPT custom MCP servers are added from Developer Mode. The labels may appear as Apps, Connectors, or New Plugin depending on the workspace surface.

  1. Open ChatGPT Settings, go to Apps & Connectors, open Advanced settings, and turn on Developer mode.
  2. Return to Apps & Connectors and choose Create app or New Plugin.
  3. Set the name to OpenADA and paste https://openada.us/mcp into Server URL. Use /mcp, not an /sse URL.
  4. For the public OpenADA demo, choose None or No authentication if that option is available. Leave OAuth credentials empty; protected deployments need their own OAuth setup.
  5. Review the custom MCP server warning, check “I understand and want to continue,” then select Create.
  6. Start a new chat, open the tools or connectors menu, enable OpenADA, and ask it to check a public page or start a scan.
https://openada.us/mcp
Open the ChatGPT submission guide
02

OpenAI Codex CLI or IDE

Add the server from Codex MCP settings, or place this configuration in ~/.codex/config.toml.

  1. Add the remote MCP server in Codex settings.
  2. Restart or refresh Codex, then approve the OpenADA tools when prompted.
[mcp_servers.openada]
url = "https://openada.us/mcp"
default_tools_approval_mode = "writes"
Read the Codex MCP guide
03

Claude custom connector

Claude uses a custom connector for remote MCP servers. Add OpenADA from the Claude settings UI, then enable it in a conversation.

  1. Open Claude Settings and choose Connectors.
  2. Select Add custom connector.
  3. Set the name to OpenADA and paste https://openada.us/mcp into Remote MCP server URL.
  4. Leave the optional OAuth Client ID and OAuth Client Secret blank for the public OpenADA demo, then select Add.
  5. Enable OpenADA in a conversation and ask it to check a public page or inspect scan history.
https://openada.us/mcp
Read the Claude MCP guide

Try it

Useful prompts

These prompts make the intended workflow clear to an AI client:

Check https://example.gov and summarize the most serious accessibility findings.Start a 50-page scan for https://example.gov, report progress, and tell me the final grade.Show the latest archived scan for www.example.gov and explain whether the score improved.

Responsible use

Built for public, reviewable work

Only public URLs are fetched. Same-host crawling is bounded and protected against private-network targets.

The MCP server is stateless; durable scan progress lives in the service so agents can reconnect and continue.

Automated scores are engineering signals, not legal advice or a certification of ADA compliance.

Need the implementation details? Read the MCP integration README or inspect the public API reference.