Use NeuroAutomata in Claude and ChatGPT with MCP

Published
producthow-tomcpclaudechatgptneuroautomata
On this page

You’re asking Claude which mutations to test next on a stability round, and to get an actual score you have to leave the conversation: open a browser tab, paste the sequence into NeuroAutomata, wait, copy the result back in. That round trip is the whole reason this connector exists. Add one URL, and Claude or ChatGPT can call NeuroAutomata’s tools directly, without you ever leaving the chat.

This guide covers what the connection is, how to set it up in each assistant, what your assistant can and can’t do once it’s connected, and how the connection itself is secured.


What this is

Claude and ChatGPT can both be extended with connectors: small bridges that let the assistant call an external tool on your behalf. NeuroAutomata publishes a connector built on the Model Context Protocol (MCP)An open standard that lets AI models connect to external tools and data sources, such as databases, APIs, and file systems, through a universal interface without custom integration code.Full definition, an open standard both assistants support. Adding it doesn’t install anything or hand over an API key. It points your assistant at NeuroAutomata and asks you to sign in.

In plain terms: a connector is like giving your assistant a library card for NeuroAutomata. The card is tied to your account, it only works after you sign in, and you can hand it back any time. Your assistant borrows the tools; it doesn’t get your NeuroAutomata login.

Because the connection runs on your account, the tools you can call and the limits on them are the same ones you’d have on the web app. Connecting doesn’t change your plan.


Why you’d want it

If you’re picking mutations by hand for the next round, ranking candidates for a stability screen, or checking whether a substitution is worth ordering an oligo for, you’re probably already talking through the design with your assistant before you touch a wet-lab plan. Right now, getting an actual ESM-2A protein language model by Meta AI trained on 250 million protein sequences. Predicts how amino acid mutations affect protein function from sequence alone — no structure required.Full definition-based score into that conversation means stepping out of it. With the connector, your assistant can generate embeddings or, on a Pro plan, rank mutations at a position or scan a whole sequence, and hand the result straight back into the thread where you’re already reasoning about the design. It’s the same scoring engine, just without the copy-paste round trip.


Before you start: create a free account

The connector signs you in with a NeuroAutomata account, so you need one first. Creating an account is free and takes a moment.

  1. Go to neuroautomata.axonagentic.ai
  2. Choose Create account and sign up
  3. Keep those credentials handy. Your assistant will ask you to sign in with them during setup.

If you try to add the connector before you have an account, the sign-in step sends you to a login screen with nowhere to go. Create the account first, then connect.


The connector URL

Both assistants ask for the same thing: this URL.

https://mcp.neuroautomata.axonagentic.ai/mcp

The /mcp at the end matters. The bare domain won’t connect. You never enter a client ID or secret alongside it; signing in with your NeuroAutomata account is the only credential the connection needs.


Connect in Claude

Claude supports custom connectors, currently a beta feature, on every plan including Free. (The Free plan allows one custom connector at a time.)

  1. Open Settings → Connectors. On a Team or Enterprise workspace, an Owner adds it under Organization settings → Connectors instead.
  2. Click +, then Add custom connector.
  3. Paste https://mcp.neuroautomata.axonagentic.ai/mcp.
  4. Click Add. Leave Advanced settings empty; no OAuth ID or secret is needed.
  5. Click Connect, sign in with your NeuroAutomata account, and approve access.
  6. In any conversation, open + → Connectors and switch NeuroAutomata on.

If Claude tells you that you already have a connector, that’s the Free plan’s one-connector limit. Remove the existing one, or upgrade to add more.


Connect in ChatGPT

ChatGPT exposes custom connectors through Developer mode, currently a beta feature, available on paid plans: Plus, Pro, Business, Enterprise, and Edu. The ChatGPT Free plan can’t add custom connectors.

If you’re on ChatGPT Free, use NeuroAutomata on the web, or connect from Claude instead, since Claude’s Free plan supports connectors.

  1. Open Settings → Apps & Connectors → Advanced settings and turn on Developer mode.
  2. Back in Apps & Connectors, click Create.
  3. Name it NeuroAutomata and paste https://mcp.neuroautomata.axonagentic.ai/mcp as the URL.
  4. Click Create, then sign in with your NeuroAutomata account and approve access.
  5. In a conversation, open + → More and select NeuroAutomata. You enable it per conversation.

If ChatGPT doesn’t show a Create option under Apps & Connectors, Developer mode isn’t switched on, or your plan doesn’t include it.


What you can run

Once connected, your assistant has five tools, gated to the same tiers as the web app.

ToolWhat it doesPlan
Generate embeddings (generate_embeddings)Mean-pooled ESM-2 embeddings for one or many sequencesFree
Check job status (get_job_status)Poll a running job and retrieve its resultFree
Score mutations (score_mutation)Effect scores for point mutations at a positionPro
Scan landscape (scan_landscape)Systematic mutational scan across positionsPro
Predict structure (predict_structure)3D structure prediction, for sequences up to 400 residuesPro

Note: These tools process the sequences you submit on infrastructure outside Axon, including a third-party service for structure prediction. What that means for your data is covered below and in our privacy policy.

A Free-tier sign-in only unlocks the two Free tools. The three Pro tools hard-reject a Free-tier sign-in; there’s no partial or silent access, and upgrading in your account is what turns them on.

In practice you don’t call these by name. You ask in plain language (“embed these three sequences,” “which mutations at position 84 look most destabilizing?”) and the assistant picks the right tool and shows you the result.

NOTE

Score mutations and scan landscape return Research Use Only (RUO)Research Use Only — a regulatory designation meaning the tool provides research scores, not clinical diagnoses. The same label used by REVEL, CADD, AlphaMissense, and PolyPhen-2.Full definition scores: relative rankings built for prioritizing which mutations to test next, not calibrated probabilities and not a stand-alone classification decision. Same positioning as the web app.


Tiers and limits

The connector enforces the same free and Pro limits as the web app; connecting doesn’t change your plan or unlock anything you don’t already have. The one difference worth knowing is concurrency: the Free tier runs up to 3 jobs at once, and Pro runs up to 10. Both caps are enforced, not advisory.


How the connection stays secure

This section is about the connection itself: who’s allowed to call the tools, and what a tool call actually carries. It is a separate question from what happens to a sequence once it reaches our server, and the next section answers that one directly.

The connector authenticates with OAuth 2.1 and PKCE (RFC 7636), the mechanism most modern apps use so one service can act on your behalf without ever seeing your password. When you approve the connection, your assistant receives a scoped access credential tied to your NeuroAutomata account, not your login credentials. You can revoke it any time from your assistant’s connector settings, and that access stops working immediately.

The connector only receives the requests you send it, not the rest of your assistant’s activity.

That’s the connection boundary, and it’s the whole of what this section covers. What NeuroAutomata does with a sequence after it arrives is a different question, answered below.


What happens to your sequences

How we handle the sequences you submit, including through this connector, is covered in one place: our privacy policy. Read it before you send anything sensitive through the connector.


Listed on Smithery, works with other MCP clients too

Claude and ChatGPT are the two most common ways in, but the connector speaks the open Model Context Protocol, so any MCP-compatible client can use it, including Cursor and Windsurf. NeuroAutomata is also listed as a server on Smithery, a registry that can walk several of these clients through setup for you.


Try it

Create a free NeuroAutomata account, then add the connector to whichever assistant you already have open. Get started. If you hit a snag connecting, reach out and we’ll help you sort it out.


TL;DR

  • One connector URL, https://mcp.neuroautomata.axonagentic.ai/mcp, adds NeuroAutomata to Claude or ChatGPT. The /mcp is required.
  • A free NeuroAutomata account is required first. No client ID or secret, just your account sign-in.
  • Custom connectors are currently a beta feature in both assistants. Claude: every plan, including Free (one connector at a time). ChatGPT: paid plans only, via Developer mode. ChatGPT Free can’t add custom connectors.
  • Five tools, tier-gated exactly like the web app: embeddings and job status on Free; mutation scoring, landscape scan, and structure prediction on Pro.
  • Free runs up to 3 concurrent jobs, Pro up to 10, both enforced.
  • The connection uses OAuth 2.1 with PKCE and a scoped access credential; the connector only receives the requests you send it, not the rest of your assistant’s activity. What happens to your sequence after that is answered in our privacy policy, not in this guide.