Connect AI Apps (MCP)¶
You can connect an AI app, such as Claude, to your ARD account. Once connected, you can ask it about your own vehicles and results, and have it do the work: read a setup, check it for warnings, make an edit, run a simulation or a sweep, and analyse what comes back.
ARD works with any app that supports the Model Context Protocol (MCP), including Claude, Claude Code and Cursor.
Availability
Connecting an AI app is available on Team+ and Enterprise plans, the same plans that include API access.
Server URL: https://api.appliedracingdynamics.com/mcp
Connecting¶
There are two ways to connect. Which one you use depends on the app.
- Sign in. The app sends you to ARD to sign in and approve it. Use this with Claude on the web and in Claude Desktop.
- API key. You paste an API key into the app's settings. Use this with Claude Code, Cursor, and apps that ask for a URL and headers.
Claude (web and desktop)¶
- In Claude, add a custom connector and give it the server URL above.
- Claude opens an ARD sign-in page. Sign in if you aren't already.
- ARD shows which app is asking for access. Choose the team it should work in, then click Allow.
You are sent back to Claude, and the connection is ready. If a team isn't offered, its plan doesn't include API access.
Claude Code¶
Generate an API key under User Settings → API Access, then run:
claude mcp add ard --transport http https://api.appliedracingdynamics.com/mcp --header "Authorization: Bearer YOUR_API_KEY"
Cursor and other apps¶
Add an MCP server that uses HTTP, with the server URL above and this header:
Some apps call the header setting headers. X-Api-Key: YOUR_API_KEY works in place of the Authorization header.
What the app can do¶
A connected app acts as you, in one team: the team you chose when signing in, or the API key's team. It can:
- Read your library: vehicles, components, tracks and ambient conditions, found by name
- Check a vehicle for setup warnings, read what the solver derives from it without a run (masses, rates, static ride heights, performance envelope), and compare two vehicles or versions field by field. It can also check a draft vehicle without saving it: whether it builds, its warnings and reports, and how it differs from a saved car
- Create a vehicle from an ARD template or from a vehicle document, and import components into your library. Identical components are linked rather than duplicated, and a clashing name is renamed or the app asks you how to resolve it
- Edit a vehicle, track or ambient conditions, including uploading a
.tirfile to a vehicle's tyres - Select the vehicle you have open in ARD, when you ask it to
- Run simulations and sweeps, using your saved presets, price a sweep before running it, wait for runs to finish, and cancel one that is still running
- Read results: lap times, metrics, swept values and channels, with your team's math channels evaluated on request, or download a result's full channel file
- Work with folders: find a folder by name, list what is in it, and download every result's channel file at once
- Search this user guide
Things to know:
- Runs spend your team's credits, exactly as they do in the app. Sweeps cost one run per combination, so a large sweep adds up quickly.
- Edits create new versions. Nothing is overwritten. Locked components can't be edited, and nor can components a bundle hides from you.
- It doesn't change what you have open in ARD unless you ask. Your selected vehicle, track and ambient conditions stay as you left them unless you ask the app to select a vehicle, and your current team never changes.
- It can only reach its own team. It can't see your other teams, even if you belong to them.
Check its work
AI apps can make mistakes. Most apps ask before using a tool that changes something, such as an edit or a run. Read what it is about to do before you approve it, and check its conclusions against your results.
Things to ask¶
- "Check my car for warnings and tell me what stands out."
- "Run a lap of Silverstone with the vehicle Test Car and tell me where the time goes."
- "Sweep the front spring rate from 80 to 120 N/mm in five steps and show me the lap times."
- "Compare my last two results and explain the difference."
Correlation report¶
The Correlation report prompt compares your ARD simulation results against reference results you uploaded to ARD, such as logged telemetry or another tool's output for the same setups. It produces a report that pairs each ARD run with its reference, compares every quantity both sides have, and lists where the two differ and what each gap most plausibly traces to.
- Put the ARD runs in one folder and upload the reference results into another. Name runs consistently on both sides (car, event, session, run), because runs are paired by name.
- In your AI app, choose the Correlation report prompt from ARD's prompts. In Claude, click + in the message box, then Connectors and ARD. In Claude Code, type
/and pick it from the list. - Enter the two folder names. Optionally add notes: quantities to focus on, tolerances, or the
summary.jsonfrom a previous report.
The report is a single HTML file you can open in any browser, with:
- the conventions resolved before comparing, such as turn direction, sign conventions and where each travel is measured from
- the register of discrepancies: each gap, its size, its most likely cause and what would settle it
- every quantity at a glance, with bias, RMS and tolerance
- reference against ARD for each quantity, one point per run
- a run explorer, showing ARD's full sweep against the reference's points for one run
- the table of which run was paired with which, and any runs left unpaired
It also writes a summary.json. Give that back next time and the report shows what changed since the last one: which gaps closed and which opened.
It needs an app that can run code
The AI app does the comparison in code, so it needs to be able to run code. Claude Code can. In Claude on the web and desktop, turn code execution on.
For full lap traces, the app also has to download the results, which Claude on the web and desktop only does when ARD's download address is allowed in its network settings. Short results, such as cornering and straight-line sweeps, don't need downloads: the app reads them directly from ARD. If the app can't get the data, it says so and offers a smaller comparison of a few runs.
Check the pairing first
The report's pairing table shows which ARD run was matched with which reference. Read it before any of the numbers. If names don't line up well, the app asks you before going on.
Limits¶
A connected app has the same limits as an API key:
- 10 requests per second, and 300 per minute
- 5 simulations running at once, counting runs you start in the app
One question can take several requests. When the app reaches a limit, it is told how long to wait. See External API: Limits.
Disconnecting¶
- Apps connected by signing in are listed under User Settings → API Access → Connected apps. Click Disconnect, and the app stops working immediately.
- Apps connected with an API key stop working when you revoke or regenerate the key.
A connection also stops working if you leave its team, or if the team's plan no longer includes API access.
Troubleshooting¶
- The app says it isn't authorised. The API key is missing, mistyped or revoked, or the team's plan doesn't include API access. Some apps report this only as a generic server error.
- Sign-in says the link has expired. Approval links last 15 minutes. Start connecting again from the app.
- "Too many requests". The app hit a limit. It can try again after a short wait.
- A run is refused for credits or plan. The app can tell you which limit applied. Top up credits or check your plan under User Settings → Billing.
Next Steps¶
- External API - The same capabilities, for your own scripts
- Ask ARD - Help with the platform itself, inside the app