Documentation | NovoMCP
Quick Start Authentication Tool Reference Resources Examples NovoMCP NovoQuantNexus

Documentation

Setup, authentication, tool reference, and enterprise connector guides.

Contents

Quick Start

NovoMCP is a remote MCP server with OAuth support that works with any MCP-compatible AI client: Claude, ChatGPT, Cursor, Windsurf, GitHub Copilot, and more.

Prerequisites: A NovoMCP API key (request access) and an MCP-compatible AI client (Claude, ChatGPT, Cursor, etc.).

Claude Desktop / Web / Mobile (GUI)

  1. Go to Settings > Connectors
  2. Click "Add custom connector"
  3. Enter URL: https://ai.novomcp.com/mcp
  4. Click Connect — a login page opens in your browser
  5. Enter your NovoMCP API key (nmcp_...)
  6. Click Authorize Access — done!

NovoMCP uses OAuth 2.0 for authentication. No config files or headers needed—just add the URL and authenticate in your browser.

Claude Code (Terminal)

  1. Add the server: claude mcp add --transport http novomcp https://ai.novomcp.com/mcp
  2. Start a session: claude
  3. Authenticate: type /mcp — browser opens for OAuth login
  4. Enter your API key and authorize — 27 tools now available!

Auth tokens are stored in your system keychain and refresh automatically on future sessions.

Team Setup (.mcp.json)

For teams, add NovoMCP to your repo's .mcp.json so every team member gets it automatically:

{ "mcpServers": { "novomcp": { "type": "http", "url": "https://ai.novomcp.com/mcp" } } }

Each team member still authenticates individually via /mcp on first use.

Cursor / Windsurf

  1. Open Settings → MCP (or equivalent)
  2. Add server URL: https://ai.novomcp.com/mcp
  3. Authenticate with your API key
  4. Done — all 27 tools available in chat

Other MCP Clients

Any MCP-compatible client can connect using:

Server URL: https://ai.novomcp.com/mcp Transport: Streamable HTTP Auth: OAuth 2.0 (automatic) or API key header

Claude Desktop (Config File Alternative)

If you prefer manual configuration:

{ "mcpServers": { "novomcp": { "url": "https://ai.novomcp.com/mcp", "headers": { "X-API-Key": "nmcp_your-api-key" } } } }

Config file locations:

  • macOS: ~/Library/Application Support/Claude/claude_desktop_config.json
  • Windows: %APPDATA%\Claude\claude_desktop_config.json

Restart Claude Desktop after saving.

Supported clients: Claude (web, desktop, mobile, terminal, API), ChatGPT, Cursor, Windsurf, GitHub Copilot, and any client that implements the Model Context Protocol.

Try it

Once connected, ask your AI:

"What's the ADMET profile of aspirin? Is it hepatotoxic?"

Your AI calls get_molecule_profile automatically and returns toxicity predictions, drug-likeness scores, regulatory status, and more. See the Examples section for full response format.

Authentication

NovoMCP supports two authentication methods:

OAuth 2.0 (Recommended)

When you add NovoMCP as a connector, your client automatically uses OAuth. You'll be redirected to a login page where you enter your API key. This is the simplest setup—no headers or config files needed.

API Key Header (Direct Access)

For programmatic access or config file setup, pass your API key in the X-API-Key header or Authorization: Bearer header.

Getting a Key

Request access via email. Keys are provisioned within 24 hours. Each key includes an organization ID and name.

Key Format

Keys follow the format nmcp_[random]. Store securely — keys cannot be retrieved after creation.

Tool Reference

NovoMCP exposes 27 tools, 5 resources, and 4 prompts via the Model Context Protocol. Your AI client invokes these automatically.

Molecular Intelligence

get_molecule_profile
Retrieve comprehensive molecular profile including properties, ADMET predictions, and regulatory compliance status.
Parameters: identifier (SMILES, CID, or name)
get_molecule_info
Basic molecular properties via RDKit: molecular weight, LogP, TPSA, hydrogen bond donors/acceptors, rotatable bonds.
Parameters: smiles
calculate_properties
On-demand RDKit property calculation: MW, exact mass, LogP, LogD, TPSA, SA Score, Lipinski/Veber violations, QED.
Parameters: smiles
get_3d_properties
32+ 3D molecular properties from conformer generation: geometry (radius of gyration), energy, electrostatics (dipole), surface/volume.
Parameters: smiles, include_coordinates
search_similar
Find structurally similar molecules using Tanimoto similarity. Returns top matches with similarity scores.
Parameters: smiles, threshold (0.0-1.0), limit
filter_molecules
Filter database by property ranges. Supports molecular weight, LogP, QED, TPSA. Exclude controlled substances.
Parameters: filters (object), limit
batch_profile
Profile multiple molecules in a single request. Up to 100 compounds per call.
Parameters: identifiers (array)
optimize_molecule
Generate optimized molecular variants using NVIDIA MolMIM. Target specific properties. All outputs screened for compliance.
Parameters: smiles, target_qed, target_logp, num_variants

ADMET & FAVES Compliance

FAVES (Fairness, Accountability, Validity, Ethics, Safety) is NovoMCP's inline compliance layer. It screens every molecule against eight regulatory jurisdictions—DEA, FDA, EPA, CWC, EU REACH—as part of the prediction, not as a separate step. Profiles, optimizations, and library screens all include FAVES results automatically.

predict_admet
40+ ML-based ADMET predictions: cardiotoxicity (1d/5d/10d/30d), CYP450 inhibition, nuclear receptors, hepatotoxicity, Ames.
Parameters: smiles
check_compliance
Full FAVES regulatory assessment with context-aware evaluation for intended use case.
Parameters: smiles, context
screen_library
Screen compound libraries up to 1000 molecules with optional full FAVES assessment and filtering.
Parameters: identifiers (array), filters, full_faves

Structure Prediction

get_protein_structure
Smart protein structure resolver with 3D visualization. Accepts protein names (EGFR, CDK2), PDB IDs (1M17), or amino acid sequences. Returns experimental structures from RCSB PDB or OpenFold3 predictions.
Parameters: target (name, PDB ID, or UniProt ID), sequence (optional), include_ligands (default true)
predict_structure
Predict 3D protein structure from amino acid sequence via OpenFold3. Async - returns job ID.
Parameters: sequence, name
get_structure_result
Retrieve completed structure predictions. Check status and download PDB files with confidence scores.
Parameters: job_id

Literature & Data Search

search_literature
Semantic search across 14,000+ curated drug discovery papers. Returns titles, authors, DOIs, relevance scores.
Parameters: query, top_k
search_patents
Search 2,400+ USPTO pharmaceutical patents. Returns patent numbers, applicants, abstracts, relevance scores.
Parameters: query, top_k
search_biorxiv
Search bioRxiv/medRxiv preprint servers for latest research. 250k+ preprints with full metadata.
Parameters: query (e.g., "BRAF inhibitors"), server (biorxiv/medrxiv), top_k, days_back
search_chembl
Search ChEMBL database for compounds, targets, and bioactivity. 2.4M compounds with SMILES and activity data.
Parameters: query (e.g., "Osimertinib"), search_type (compound/target/activity), top_k
search_clinical_trials
Search ClinicalTrials.gov for clinical studies. 500k+ trials with status, phase, and enrollment.
Parameters: query (e.g., "obesity treatment"), status, phase, top_k

Tip: Use search_literature, search_patents, search_biorxiv, and search_clinical_trials together to build a comprehensive landscape analysis for a specific target or compound.

Platform & Account

get_platform_info
Platform information: subscription tiers, database statistics, ADMET capabilities, compliance lists, usage summary.
Parameters: info_type (all, tiers, database, admet, compliance, usage)
get_job_status
Check status of async jobs (structure prediction, molecular dynamics).
Parameters: job_id
get_credit_usage
Check your credit balance, usage statistics, and research value realized. Returns tier info, usage percentage, and interactive dashboard.
Parameters: type (summary or detailed)
list_connections
List configured data export connections for your organization. Shows connection status and available destinations.
Parameters: none

Data Connectors

pull_from_source
Bidirectional data pipeline: pull compounds from Databricks/Snowflake, run ADMET + FAVES processing, push enriched results back. 4 actions: preview, pull, estimate_pipeline, execute_pipeline.
Parameters: action (preview/pull/estimate_pipeline/execute_pipeline), connection_id, table, filters, processing_tools, confirmation_token
discover_schema
Discover target system tables, columns, and types. Supports all 4 connector adapters with normalized type mapping.
Parameters: connector_type, connection_id
preview_mapping
Preview field mapping between NovoMCP tool output and target connector schema. Supports template, auto, and AI-assisted mapping strategies.
Parameters: tool_name, connector_type, target_table
export_results
Write molecular intelligence data to an external system. Snowflake, BigQuery, Databricks, Supabase.
Parameters: connector_type, connection_id, data, target_table

Resources

NovoMCP exposes 5 resources that provide static and dynamic data. Access via GET /mcp/resources.

compliance_schedules
DEA Schedule I-V substances, CWC chemical weapons lists, FDA banned substances, EPA PBT chemicals, and EU REACH restricted compounds.
admet_properties
List of 40+ ADMET property predictions available: absorption, distribution, metabolism, excretion, and toxicity endpoints.
tier_features
Features available at each subscription tier with tool access details.
database_stats
Current database statistics: 122M indexed molecules, 14,398 curated papers, 2,416 USPTO patents. Updated periodically.
changelog
Recent changes to NovoMCP tools, features, and API. Check this if tools seem missing—you may need to reconnect your MCP connection to refresh the tool list.

Prompts

NovoMCP provides 4 pre-defined prompts for common workflows. Access via GET /mcp/prompts.

quick_check
Quickly check if a molecule is a controlled substance or has safety flags. Returns compliance status and any alerts.
Arguments: smiles (required)
full_analysis
Comprehensive molecular analysis including properties, ADMET predictions, compliance status, and drug-likeness scores.
Arguments: smiles (required)
find_alternatives
Find structurally similar molecules that pass compliance checks. Useful when a lead compound is flagged.
Arguments: smiles (required), min_similarity (optional, default 0.7)
literature_review
Search drug discovery papers and patents for a topic or target. Returns relevant literature with citations.
Arguments: query (required), max_results (optional, default 10)

Enterprise Connectors

Export molecular intelligence data directly to your data warehouse or collaboration tools. 4 connector adapters with OAuth sign-in, dynamic schema discovery, AI-assisted field mapping, and full governance.

Available Connectors

  • Snowflake
  • Databricks
  • BigQuery OAuth
  • Supabase

How It Works

The export workflow follows three steps:

  1. Discover: discover_schema introspects the target system—tables, columns, types—and normalizes everything to 5 standard types (string, number, boolean, datetime, json).
  2. Map: preview_mapping aligns NovoMCP tool output fields to target columns. Uses template mapping for common combos, auto-matching by name, or AI-assisted mapping via Azure OpenAI with confidence scores.
  3. Export: export_results writes data to the target system through the appropriate connector adapter.

Governance

Enterprise exports are secured by novomcp-auth—an identity and governance service implementing RFC 8693 token exchange. Your nmcp_ API key is exchanged for a scoped JWT with connector-specific permissions. All export operations are audit-logged with full traceability.

Credentials: All connector credentials are stored in Azure Key Vault—never in the database. Each connector supports test_connection to verify credentials before any data export.

Connector Setup Guides

Each connector has its own authentication method. Follow the guide for your target system.

BigQuery OAuth

Click "Connect with Google" in the connections dashboard. You'll be redirected to Google's consent screen to authorize BigQuery access. No API keys or service accounts needed—your Google account credentials are used directly. Tokens refresh automatically.

Alternative: You can also connect using a Google Cloud service account JSON key for server-to-server access without user interaction.

Snowflake

Provide your Snowflake account identifier (e.g. xy12345.us-east-1), username, and password. You'll also need to specify a warehouse, database, and schema in the connection config.

Find your account identifier in Snowflake under Admin > Accounts.

Databricks

Generate a personal access token in Databricks: User Settings > Developer > Access Tokens > Generate New Token. Provide the token along with your workspace URL (e.g. https://adb-1234567890.1.azuredatabricks.net) and target catalog/schema.

Supabase

Find your connection details in your Supabase project dashboard under Settings > Database > Connection String. Use the service_role key for full table access, or the anon key for RLS-restricted access.

Examples

Ask in natural language. NovoMCP tools are selected and invoked automatically—no tool names or parameters needed.

Basic Profiling

"What's the ADMET profile of ibuprofen? Is it hepatotoxic?"

Your AI invokes get_molecule_profile with identifier "ibuprofen" and receives:

Response
{ "properties": { "cid": 3672, "smiles": "CC(C)CC1=CC=C(C=C1)C(C)C(=O)O", "molecular_weight": 206.28, "logp": 3.97, "tpsa": 37.3, "qed": 0.72 }, "admet": { "hepatotoxicity_probability": 0.12, "cardiotoxicity_probability": 0.08, "bbb_permeant": true, "cyp2d6_inhibitor": false, "human_intestinal_absorption": 0.98 }, "compliance": { "status": "clear", "is_dea_controlled": false, "is_fda_banned": false, "structural_alerts": [] } }

Additional Examples

Each query maps to specific tools, selected automatically based on your prompt.

"Optimize this lead compound for better oral bioavailability while maintaining selectivity: CC1=CC=C(C=C1)C(=O)NC2=CC=CC=C2"

→ optimize_molecule — generates variants targeting QED and LogP thresholds

"Is this molecule safe to develop? Check for any regulatory concerns: CN1C=NC2=C1C(=O)N(C(=O)N2C)C"

→ get_molecule_profile — returns DEA, FDA, structural alert status

"What's known about CDK4/6 inhibitors in breast cancer? Show me literature and clinical trials."

→ search_literature + search_clinical_trials — parallel search across sources

"Find molecules similar to palbociclib with QED > 0.6 and no PAINS alerts"

→ search_similar — similarity search with property and compliance filters

"Pull imatinib from my Databricks compound_library. Run ADMET and FAVES compliance. Push enriched results back to admet_enriched_leads."

→ pull_from_source — preview → estimate → execute pipeline

Combining search tools: search_literature, search_patents, search_biorxiv, and search_clinical_trials can be combined in a single prompt for landscape analysis. Example: "Give me a complete landscape analysis of KRAS inhibitors."

Novel molecules: Compounds not in the pre-computed database are processed on-the-fly via RDKit with FAVES screening. Response times: ~500ms vs ~50ms for indexed compounds.