Connect your AI tools

Give Claude, Cursor, ChatGPT or any MCP client direct access to the Aeron documentation.

This documentation is available to AI tools as an MCP server. Point your assistant at it once and it can search these pages and read them directly, instead of you copying text into a chat window or hoping it guessed right.

It works with Claude Code, Claude.ai, Cursor, VS Code, ChatGPT and anything else that speaks the Model Context Protocol.


Endpoint	`https://docs.aeronsystems.com/api/mcp`
Transport	HTTP (no local install, nothing to run)
Auth	None for public docs. A personal token for internal sections.
Cost	Free

Connect

1. Public documentation

Works straight away, no account needed. Reaches every public page.

claude mcp add --transport http aeron-docs /api/mcp

2. Internal documentation (Aeron accounts)

Generate a personal token to reach the sections your account has been granted. The token identifies you; what it can read is checked on every request, so access granted or revoked later takes effect immediately.

What your assistant can do

Once connected, four tools become available to it. You do not call these yourself; the assistant picks whichever it needs.

Tool	What it does
`aeron_docs_search`	Searches the documentation and returns the best matching passages.
`aeron_docs_get_page`	Reads one or more full pages.
`aeron_docs_list_sections`	Lists every section and page it is allowed to see.
`aeron_docs_status`	Reports which documentation your token can reach.

Try asking it things like:

"How do I create a Live3 API key?"
"What are the rate limits on the Live3 data API, and how should I handle a 429?"
"Write me a Python script that pulls yesterday's readings for one station."

The assistant will search these pages and answer from them, with links.

Setting it up in each tool

Run this in your terminal:

claude mcp add --transport http aeron-docs https://docs.aeronsystems.com/api/mcp

To reach internal documentation, generate a token above and add it:

claude mcp add --transport http aeron-docs https://docs.aeronsystems.com/api/mcp --header "Authorization: Bearer aeron_docs_..."

Check it worked with /mcp inside a Claude Code session. You should see aeron-docs listed as connected.

Flags go before the server name. claude mcp add aeron-docs --transport http ... will not work.

Open Settings → Connectors → Add custom connector, then paste:

https://docs.aeronsystems.com/api/mcp

Connectors added on Claude.ai also become available inside Claude Code when you are signed in with the same account.

Claude.ai cannot currently send a personal token, so it reaches the public documentation only. Use Claude Code, Cursor or VS Code for internal sections.

Add this to ~/.cursor/mcp.json (or .cursor/mcp.json in a project):

{
  "mcpServers": {
    "aeron-docs": {
      "url": "https://docs.aeronsystems.com/api/mcp"
    }
  }
}

For internal documentation, add your token:

{
  "mcpServers": {
    "aeron-docs": {
      "url": "https://docs.aeronsystems.com/api/mcp",
      "headers": { "Authorization": "Bearer aeron_docs_..." }
    }
  }
}

Add this to .vscode/mcp.json in your workspace:

{
  "servers": {
    "aeron-docs": {
      "type": "http",
      "url": "https://docs.aeronsystems.com/api/mcp"
    }
  }
}

Add a "headers" object with your Authorization bearer token to reach internal sections.

Any MCP client that supports a remote HTTP server will work. Give it:

URL — https://docs.aeronsystems.com/api/mcp
Transport — HTTP (sometimes called "streamable HTTP" or "remote server"). This is not an SSE server and there is nothing to install locally.
Header (optional) — Authorization: Bearer <your token>

Two clients worth naming. Windsurf uses the key serverUrl rather than url, which is a common copy-paste mistake. ChatGPT supports MCP servers in developer mode; add it as a connector by URL.

What you can see

The documentation has three levels, and the server shows you exactly what you would see signed in to this site in a browser. Nothing more.

Level	Contents	Who
Public	Product documentation: concepts, the interface guide, how-to guides, the data API reference	Everyone, no token needed
Engineering	Technical reference for a product: architecture, data ingestion, security, operations	Aeron staff and scoped collaborators
Internal	Company-wide internal tooling and registers	Aeron staff

Access follows your account, not your token. If someone grants or revokes your access, it takes effect on your very next question. You do not need a new token.

Pages you cannot read are not hidden behind a locked door, they simply are not there: your assistant is told they do not exist, exactly as the website would. So if you know a page exists and your assistant cannot find it, you probably need access rather than a better search query.

Your token is a password

A personal token reads everything your account can read. Keep it out of shared machines, screenshots and committed files. If one leaks, tell an administrator.

Troubleshooting

The assistant cannot find an internal page. Ask it to run aeron_docs_status. That reports who you are authenticated as and how many pages you can reach. If it says you are not authenticated, your token is not being sent. If it says you are authenticated but hold no grants, ask an administrator for access.

Connection fails immediately. Check the URL ends in /api/mcp and that your client is set to HTTP, not stdio or SSE.

"406 Not Acceptable". Your client is not sending the right Accept header. This is a bug in the client, not in your setup. Verify the server is healthy with:

curl -sX POST https://docs.aeronsystems.com/api/mcp \
  -H 'Content-Type: application/json' \
  -H 'Accept: application/json, text/event-stream' \
  -d '{"jsonrpc":"2.0","id":1,"method":"tools/list"}'

That should return a list of the four tools.

The assistant gives stale answers. Most clients cache the tool list for a session. Restart it.

Notes

Everything here is read-only. Nothing your assistant does through this server can change the documentation. Edits still go through a pull request that a person reviews and merges.

Prefer plain files? /llms.txt lists every page and /llms-full.txt returns all of it as one document. Both respect the same access rules. They are useful when you want to hand a file to a tool, but the MCP server is better for day-to-day use because the assistant fetches only what it needs.