Connect Avala to Claude, ChatGPT, Codex, Cursor, Grok, and other AI assistants in under a minute
Avala has a hosted MCP server. Add one URL to your AI assistant, sign in with your Avala account, and you can ask it about your datasets, projects, captures, quality results, and exports in plain language.
The only thing you need
https://mcp.avala.ai/mcpPaste this URL wherever your assistant asks for an MCP server or custom connector. When prompted, sign in with the same account you use for Mission Control. No API key, no install.
Open Customize → Connectors (Team and Enterprise admins: Organization settings → Connectors).
Choose Add custom connector.
Name it Avala and enter https://mcp.avala.ai/mcp.
Click Add, then Connect and sign in with your Avala account.
Avala tools now appear in the + menu of every chat.
Run once in your terminal:
claude mcp add --transport http avala https://mcp.avala.ai/mcp
Then, inside Claude Code, type /mcp and choose Authenticate next to avala. Your browser opens the Avala sign-in page; after you approve, Claude Code is connected.
In ChatGPT, open Settings → Connectors (Developer mode must be enabled on your plan).
Choose Create (or Add connector), name it Avala, and enter https://mcp.avala.ai/mcp.
Save, then Connect and sign in with your Avala account.
Enable the Avala connector from the + menu in a chat.
Complete the Avala sign-in if Codex opens your browser. If adding the server did not complete authentication, run:
codex mcp login avala
After signing in, use /mcp in Codex to check the connection and available tools. Your Avala account’s permissions determine which tools you can use.The Codex CLI, IDE extension, and ChatGPT desktop app share MCP configuration on the same Codex host. Cloud environments need separate configuration and authentication verification. See the Codex MCP documentation for setup details.
Add this to .cursor/mcp.json in your project (or your global Cursor MCP settings):
Enter https://mcp.avala.ai/mcp and sign in with your Avala account when prompted.
Any assistant that supports remote MCP servers over Streamable HTTP with OAuth works the same way: enter https://mcp.avala.ai/mcp and sign in. Examples: Windsurf (~/.codeium/windsurf/mcp_config.json, "serverUrl"), Gemini CLI ("httpUrl"), and Claude Managed Agents.For automation, CI, or a client that cannot open a browser, use an Avala API key instead of signing in — send it as X-Avala-Api-Key or Authorization: Bearer <key> on every request. Most clients let you set a header on a remote server entry.
Signing in only grants the assistant read access to what your Avala account can already see. Nothing is created, changed, or deleted through the hosted server. Your sign-in can be revoked at any time from your Avala account or from the assistant’s connector settings.
Which capture campaigns are running for the Warehouse Pick dataset, and how far along are they?
Show the acceptance summary for last week's captures — what is getting rejected and why?
Explain why capture 8f3c… was quarantined
What are the open annotation issues on the Lidar QA project?
Show my recent exports and whether they finished
The assistant picks the right Avala tool, calls it, and answers with your data. You can chain questions in one conversation — list datasets, pick one, dig into its captures, then check an export.
Sign in to Mission Control once with the same account, then try Connect again. Your Avala account must be active and its email verified. If it still fails, ask your organization admin to confirm your account has access.
The tools list is empty
Your account’s permissions decide which tools appear. Datasets you cannot see in Mission Control are not visible to the assistant either. If you expect more, ask your organization admin about your role.
My assistant asks for a client ID and secret
Leave them blank — Avala supports automatic registration, so no client credentials are needed. If your organization prefers a fixed OAuth client, your admin can request one from support@avala.ai.
Opening https://mcp.avala.ai/mcp in a browser shows 'POST only'
That is expected. The URL is for AI assistants, not browsers. Paste it into your assistant’s connector settings instead.
