How to get MARA, Nanome's AI copilot, and what it can and can't run

Nanome's AI copilot, MARA, is built into the Nanome web app at app.nanome.ai, so getting it means signing in there with a Nanome account, and no seat needs a separate MARA purchase. The free Starter seat comes with 50 MARA credits. Academic Full seats and commercial seats add the 300+ tool library, custom tools written in Python, HTTP or SQL, and, on Full Web and Full seats, 1,000 credits a month. Docking runs on Smina, AutoDock Vina and DiffDock-L. The Smina tools take one ligand or one multi-ligand SDF per run, and a screen of thousands of compounds is better sent to a GPU docking engine on a cluster. For MARA inside a company's own network, with SSO and audit logs, there's MARA Enterprise.

The old mara.nanome.ai address redirects to app.nanome.ai. New accounts get 14 days on a Full license trial, and no credit card is asked for.

What MARA does once you're in, from co-folding to ADMET, is the subject of an AI copilot for drug discovery workflows.

Where MARA runs

In the web app, MARA is a chat. It plans a workflow from a plain-English request, and since Web App 1.0 a slash menu in the chat input runs any single tool directly. Structures open in a Mol*-based viewer inside the chat, where "hide waters" or "color residues 101-207 blue" changes the view.

Each answer shows which tools ran, with what inputs and what came out. The way plain-English requests map onto tools is covered in natural language interfaces for molecular modeling.

In the XR app, Nanome v2.6 put MARA tools in the in-headset sidebar, in VR or on desktop. Per the release notes, a tool runs on a whole entry, a component or the current selection, and its results come back as new structures. The v2.6 release post calls the feature a beta.

Version 2.4.0 brought back voice commands through MARA. They handle representation changes (colors, rendering styles, visibility), including on standalone headsets such as Meta Quest, and the pricing page lists them with the Full seat. Naming the chain and the structure works better than "make this blue", since voice doesn't read gaze or gesture yet.

What each plan includes

The pricing page sets out what each seat gets. Its feature comparison gives Full Web and Full seats 1,000 credits a month, with chats and MARA workflows included, while a Starter or Collab seat has 50 credits and draws chats and workflows from them. External tools and autonomous agents are credit-based on every seat. More credits can be bought on top of any seat type and pay for tool calls, agentic use and MCP calls.

Docking: engines, scores and scale

Docking requests in MARA run on Smina, AutoDock Vina and DiffDock-L. On the integrations page's tool list, Run Smina Docking is the main tool for single-ligand runs, and Smina Multi-Ligand Docking takes a multi-ligand SDF.

A pocket finder sets the search space. Find Pockets with P2Rank returns pocket coordinates, and Compute Docking Box draws a box around chosen residues. The Flare integration walkthrough follows the same pocket-first sequence.

Smina and Vina report each pose with an estimated affinity in kcal/mol. That figure is a docking score, useful for ranking poses against each other, and a measured Kd still comes from the assay.

Evaluate Docking Poses checks every pose for hydrogen bonds and clashes. Poses open in MARA's viewer, and in a headset, multi-frame playback steps through them one at a time. Software for molecular docking visualization covers that review step in more depth.

The Smina tools take one ligand, or one multi-ligand SDF, per run. A screen across thousands of compounds fits a GPU engine such as AutoDock-GPU or Uni-Dock on a cluster, with the top hits brought back into Nanome for review. The pricing page lists token-based GPU tools for Full Web and Full seats as coming soon. A proprietary scoring function can come in as a custom tool, which the integrations page notes enterprise teams have done.

Protein prep: what's covered today

Tools on the integrations page, most of them in its Structure Cleaning & Preparation category, handle the usual first pass:

• Solvent and ions out. Save Protein-Only PDB drops ligands, solvent and ions, and can add hydrogens.
• A single chain. Split Chain from PDB writes one chain to its own file, and Delete Chain removes one.
• Repairs. PDB File Fixer runs pdb4amber, reduce and PDBFixer on the protein.
• Ligands as SDF. Download Ligand SDFs from RCSB pulls every ligand in an entry, and Prepare Ligands at pH, from SDF or from SMILES, adds hydrogens at a target pH and optimizes the geometry.
• Minimization. Minimize Protein-Ligand Complex minimizes a ligand in its protein with OpenMM, OpenFF 2.1.0 and Amber ff14SB. In the headset, the Builder minimizes a ligand with MMFF94, MMFF94s or UFF.

A system headed for MD or free-energy work needs deliberate choices about protein protonation states, alternate conformations and charges. Those usually get made in a dedicated prep tool. For teams with a Flare license, the Flare integration demo shows MARA handing protein preparation to Cresset's Flare and carrying the prepared file through pocket detection and docking.

Custom tools

A custom tool in MARA is backed by a Python snippet, for which MARA spins up a managed sandbox, an HTTP endpoint that turns any REST API into a tool, or a SQL query against a database. Tools can also be written directly as JSON, and the introduction to MARA tool creation walks through naming, arguments, code and testing.

A tool can stay personal while it's tested and then go to the whole organization. From there it chains into saved workflows like any built-in tool.

A PubChem lookup. An HTTP endpoint tool with one argument, name, pointed at PubChem's PUG REST service:

https://pubchem.ncbi.nlm.nih.gov/rest/pug/compound/name/{name}/property/MolecularFormula,MolecularWeight,SMILES,XLogP/JSON

Asking for imatinib returns CID 5291 with its formula, weight, SMILES and XLogP. MARA already ships PubChem search tools, so this serves as a pattern for any public REST API, RCSB's included.

A minimization energy. A Python snippet tool with one argument, smiles, built on RDKit. The sketch assumes the sandbox has RDKit and hands the argument to the code as a variable, and MARA's built-in AI code help can adjust it to the tool's actual inputs:

from rdkit import Chem
from rdkit.Chem import AllChem

mol = Chem.AddHs(Chem.MolFromSmiles(smiles))
AllChem.EmbedMolecule(mol, randomSeed=42)
props = AllChem.MMFFGetMoleculeProperties(mol)
ff = AllChem.MMFFGetMoleculeForceField(mol, props)
ff.Minimize(maxIts=2000)
print(f"MMFF94 energy: {ff.CalcEnergy():.2f} kcal/mol")

The Python plugin API and its NTS relay server belong to Nanome Classic, v1.24 and below. For v2, third-party integrations run through MARA.

Code reaches MARA and Nanome three ways. Every MARA tool is a REST endpoint (POST /api/tools/<tool_id>/run), called with an API key created under the System tab in Settings (technical docs), and the same key works from Jupyter. The Workspace API builds Nanome workspaces from code. Nanome also publishes a Claude Code skill and MCP server that let a coding agent create workspaces and reach the MARA tool library.

Dedicated instances

Commercial Cloud gives each customer its own Nanome-hosted stack and its own server for custom tools. Enterprise moves the deployment into the customer's own private cloud or data center and adds internal database integration, custom or internal LLM and MCP servers, and custom development.

On Enterprise the model is the customer's choice, a privately hosted or in-house model included, and the MARA page says Nanome cannot access an Enterprise customer's data.

MARA Enterprise lists SSO, organization-wide permissions for tools and queries, and audit logs with alerts. Sign-in uses SAML with Azure AD, Okta or another SAML provider, and admins can set organization-wide API keys for specific tools (technical docs).

A few setup details are worth settling early:

• Model keys. When MARA runs on the customer's own model account, that account's limits apply. An OpenAI key with no credit left, or past its spend limit, returns error 429, which shows up as failed model calls.
• Versions. A dedicated deployment pins a matching Workspace API and MARA version, so headsets run the Nanome build that matches it, listed on the versions page.
• Network. A tool that reaches an internal database or service needs a route through the firewall, and Nanome's network and security requirements lists what IT needs to allow.
• Sharing tool JSON. Tool definitions travel best as an attached file. Email security gateways that rewrite links can alter an endpoint URL pasted into a message body.

FAQ

Is MARA a separate purchase?
No. It's part of the Nanome web app and every seat reaches it: Starter with 50 credits, academic Full seats and commercial seats with the full tool library and, on Full Web and Full seats, 1,000 credits a month. Extra credits can be bought with any seat type.

Does MARA work in VR?
Yes. Since v2.6, MARA tools run from the in-headset sidebar (a beta at release), and on a Full seat, voice handles representation changes. MARA's chat lives in the web app, and its results land in Nanome workspaces that open in XR.

Does MARA work in languages other than English?
Nanome documents MARA in English, and tool names and parameters are in English too. The docs don't list other supported languages.

Where did my old Nanome plugin go?
Classic plugins run on Nanome Classic v1.24 and below. In v2, most of their jobs moved into Nanome or MARA: v2.4.0 rebuilt chemical interactions in-house and added ligand minimization, and Smina docking and APBS electrostatics are MARA tools. A custom plugin's logic can come across as a MARA custom tool.