How do I use Nixtla MCP Server?

1. Click on "Deploy Server". 2. Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state. 3. In the chat, type @ followed by the MCP server name and your instructions, e.g., "@Nixtla MCP Server forecast my daily sales for the next 30 days" That's it! The server will respond to your query, and you can continue using it as needed. Here is a step-by-step guide with screenshots.

Nixtla MCP Server

by VaclavRut

Overview Schema Source Related Servers Discussions Admin

TypeScript

Remote

Nixtla MCP Server

MCP server for multitenant Nixtla TimeGPT forecasting with fine-tuning, rolling backtests, and comprehensive usage tracking.

Features

🔐 Multitenant Auth: Token-based authentication with organization isolation
🎯 Fine-tuning: Train and manage organization-specific models with metadata
🤖 Smart Base Model Selection: Automatically selects timegpt-1 or timegpt-1-long-horizon based on data characteristics
📊 Rolling Backtest: Prove fine-tuned models outperform baseline
🚨 Anomaly Detection: Detect anomalies with baseline or finetuned models
📈 Usage Tracking: Granular consumption tracking with performance metrics
🌐 HTTP MCP: JSON-RPC 2.0 over HTTPS (not stdio)
🏢 Unified Organization Management: Models, tokens, and configs in one place
💾 Blob Storage: Secure 30-day TTL storage with organization-based access control
🔮 X_future Support: Use future exogenous variables for better forecasts

Related MCP server: forecast-mcp

Quick Start

1. Deploy to Vercel

Deploy with Vercel

Or manually:

# Clone the repository
git clone https://github.com/VaclavRut/nixtla-mcp.git
cd nixtla-mcp

# Install dependencies
npm install

# Deploy to Vercel
vercel

2. Set Environment Variables

In your Vercel project settings, add these environment variables:

# Nixtla API Key (get from https://dashboard.nixtla.io)
NIXTLA_API_KEY=your_nixtla_api_key_here

# HMAC secret for hashing authentication tokens (generate a strong random string)
MCP_AUTH_TOKEN_SECRET=your_strong_random_secret_here

# Admin token for bootstrap endpoint (generate a strong random string)
ADMIN_SETUP_TOKEN=your_admin_token_here

# MongoDB connection string (MongoDB Atlas or your own MongoDB instance)
MONGODB_URI=mongodb+srv://username:password@cluster.mongodb.net/?retryWrites=true&w=majority

# Vercel Blob storage token (for storing forecast/anomaly results)
BLOB_READ_WRITE_TOKEN=your_vercel_blob_token_here

Generating Secure Secrets:

# Generate random secrets (run these in your terminal)
node -e "console.log(require('crypto').randomBytes(32).toString('base64'))"

3. Set Up MongoDB

You have two options:

Option A: MongoDB Atlas (Recommended - Free tier available)

Go to MongoDB Atlas
Create a free cluster
Create a database user
Whitelist Vercel's IP addresses (or use 0.0.0.0/0 for all IPs)
Get your connection string and add it to MONGODB_URI

Option B: Self-hosted MongoDB

# Install MongoDB locally or use Docker
docker run -d -p 27017:27017 --name mongodb mongo:latest

# Connection string for local MongoDB
MONGODB_URI=mongodb://localhost:27017/nixtla-mcp

4. Set Up Vercel Blob Storage

Go to your Vercel project dashboard
Navigate to Storage → Create Database → Blob
Copy the BLOB_READ_WRITE_TOKEN and add it to environment variables

5. Bootstrap Your First Organization

After deploying, create your first organization:

curl -X POST https://your-deployment.vercel.app/api/admin/bootstrap-company \
  -H "Authorization: Bearer YOUR_ADMIN_SETUP_TOKEN" \
  -H "Content-Type: application/json" \
  -d '{
    "orgId": "my-organization"
  }'

Response:

{
  "orgId": "my-organization",
  "authToken": "abc123xyz...",
  "message": "Organization bootstrapped successfully. Store the authToken securely - it will not be shown again."
}

⚠️ Important: Save the authToken - it's returned only once and is needed for all API calls!

Architecture

AI Agent (Claude, GPT, etc.)
    ↓ HTTP POST with Bearer token
MCP Server (Vercel)
    ↓ Auth via MongoDB → Organization resolution
    ↓ Load dataset from URL or inline data
    ↓ Call Nixtla TimeGPT API
    ↓ Track granular usage → MongoDB
    ↓ Store results → Vercel Blob (30-day TTL)
    ↓ Return JSON-RPC response with download URLs

Dataset Format

Datasets can be provided via datasetUrl (URL or blob storage path) or datasetData (inline JSON) in Nixtla multi-series format:

{
  "series": {
    "y": [100, 105, 110, 108, 112, 115],
    "sizes": [3, 3],
    "X": [[1.0, 2.0], [1.1, 2.1], [1.2, 2.2], [1.3, 2.3], [1.4, 2.4], [1.5, 2.5]],
    "X_future": [[1.6, 2.6], [1.7, 2.7]]
  },
  "freq": "D",
  "meta": {
    "seriesNames": ["series_a", "series_b"],
    "timestamps": ["2024-01-01", "2024-01-02", "2024-01-03"]
  }
}

y: Concatenated values across all series (historical)
sizes: Length of each series (must sum to y.length)
X: Optional exogenous variables (historical, same length as y)
X_future: Optional future exogenous variables for forecast period
freq: D=Daily, H=Hourly, W=Weekly, M=Monthly, MS=Month Start

MCP Tools

Authentication

All MCP requests require:

POST https://your-deployment.vercel.app/mcp
Authorization: Bearer <YOUR_AUTH_TOKEN>
Content-Type: application/json

Available Tools

1. `validate_dataset`

Validate dataset format before forecasting.

2. `finetune_model`

Fine-tune a model on your data with automatic base model selection:

{
  "jsonrpc": "2.0",
  "id": "3",
  "method": "tools/call",
  "params": {
    "name": "finetune_model",
    "arguments": {
      "datasetUrl": "https://blob.vercel-storage.com/...",
      "finetuneOptions": {
        "model": "timegpt-1-long-horizon",  // Optional: override auto-selection
        "finetune_steps": 300,
        "finetune_loss": "mae",
        "finetune_depth": 4,
        "output_model_id": "my-custom-model"
      }
    }
  }
}

Smart Base Model Selection:

Auto-selected by default based on average series length
Manual override available via model parameter
Automatically runs rolling backtest to validate improvement

3. `forecast`

Generate forecasts using your organization's active model or baseline:

{
  "jsonrpc": "2.0",
  "id": "4",
  "method": "tools/call",
  "params": {
    "name": "forecast",
    "arguments": {
      "datasetUrl": "https://blob.vercel-storage.com/...",
      "h": 14,
      "level": [80, 95],
      "feature_contributions": true
    }
  }
}

Returns: JSON and CSV download URLs (30-day expiration)

4. `anomaly_detect`

Detect anomalies in time series data:

{
  "jsonrpc": "2.0",
  "id": "5",
  "method": "tools/call",
  "params": {
    "name": "anomaly_detect",
    "arguments": {
      "datasetUrl": "https://blob.vercel-storage.com/...",
      "useFinetunedModel": true,
      "params": {
        "level": 99,
        "clean_ex_first": true
      }
    }
  }
}

5. `list_org_models`

List all fine-tuned models for your organization.

Connecting AI Agents

Claude Desktop

Add to your Claude Desktop config (~/Library/Application Support/Claude/claude_desktop_config.json on macOS):

{
  "mcpServers": {
    "nixtla": {
      "url": "https://your-deployment.vercel.app/mcp",
      "transport": "http",
      "headers": {
        "Authorization": "Bearer YOUR_AUTH_TOKEN"
      }
    }
  }
}

Custom Integration

import requests

MCP_SERVER_URL = "https://your-deployment.vercel.app/mcp"
AUTH_TOKEN = "your-auth-token"

response = requests.post(
    MCP_SERVER_URL,
    headers={"Authorization": f"Bearer {AUTH_TOKEN}"},
    json={
        "jsonrpc": "2.0",
        "id": "1",
        "method": "tools/call",
        "params": {
            "name": "forecast",
            "arguments": {
                "datasetUrl": "https://blob.vercel-storage.com/...",
                "h": 14
            }
        }
    }
)

print(response.json())

Development

# Install dependencies
npm install

# Run type checking
npm run build

# Run locally (requires environment variables)
vercel dev

Security Features

Hashed tokens: Raw tokens never stored, only HMAC hashes
Organization isolation: All data scoped by orgId with MongoDB indexes
Permission system: Admin vs user access controls
Idempotency: Prevents duplicate usage charges via request IDs
Secure blob storage: 30-day TTL with organization-scoped access

License

MIT

Contributing

Contributions are welcome! Please open an issue or submit a pull request.

Support

Issues: GitHub Issues
Documentation: Full API Reference
Nixtla Docs: https://docs.nixtla.io

This server cannot be deployed

One endpoint for every MCP toolRayrun runs every MCP server behind one URL and holds the credentials. Each call is decided per client, tool, and argument, then recorded.ray.runAd

Maintenance

ActivityInactive

ResponsivenessNo issues

Resources

GitHub Repository

Need Help?

Related Servers

Related MCP Connectors

EvalGate B
io.github.evalgate
Create and run AI evaluations, record traces, and inspect results with scoped organization access.
AI & Machine Learning Testing & QA Tools Observability
121
WeltOAuth
dev.ergodic.welt
Run Welt's tabular, causal and forecasting models on your data. Hosted; OAuth sign-in.
AI & Machine Learning Research & Data
1
Argus API Spend B
town.argus
Track API spend, credit balances and runway across OpenAI, Anthropic, OpenRouter and 25 more.
AI & Machine Learning Finance Monitoring
121
agent-cost-allocator-mcp
io.github.CSOAI-ORG
Agent Cost Allocator MCP — multi-tenant LLM cost attribution for chargeback billing. Companion to
Finance Payments & Billing
2

Related MCP Servers

shopify-forecast-mcp
E-commerce & Retail AI & Machine Learning
mcostigliola321
A
license
A
quality
C
maintenance
Enables AI assistants to forecast revenue and demand for Shopify stores using Google's TimesFM model. Provides tools for revenue forecasting, demand analysis, promotion analysis, and anomaly detection.
Updated 6 months ago (2026-04-20 16:39 UTC)
7
1
MIT
forecast-mcp
AI & Machine Learning Research & Data
ramdhavepreetam
A
license
B
quality
D
maintenance
Enable any AI agent to forecast time-series data (e.g., sales, traffic) using Google's TimesFM or a zero-dependency statistical baseline.
Updated 5 months ago (2026-06-02 00:50 UTC)
3
Apache 2.0
data-science-mcp
LiChenStuttgart
A
license
Not graded
quality
D
maintenance
Enables time-series analysis and forecasting through a structured tool catalogue, including data loading, quality repair, diagnostics, and forecasting with ARIMA, exponential smoothing, Chronos-2, Toto 2.0, and AutoML.
Updated 3 months ago (2026-07-20 14:48 UTC)
1
MIT
TimesFM-3 Studio MCP Server
AI & Machine Learning Research & Data
TheAIHorizon
A
license
Not graded
quality
B
maintenance
Enables AI agents to run TimesFM-3 forecasting workflows locally, including joint multivariate forecasting with known future drivers, backtesting against a baseline, what-if scenario comparison, and historical anomaly detection. It exposes the studio's tools and bundled public and synthetic demo datasets over MCP.
Updated a month ago (2026-09-13 11:32 UTC)
Apache 2.0

Nixtla MCP Server

Features

Quick Start

1. Deploy to Vercel

2. Set Environment Variables

3. Set Up MongoDB

4. Set Up Vercel Blob Storage

5. Bootstrap Your First Organization

Architecture

Dataset Format

MCP Tools

Authentication

Available Tools

1. validate_dataset

2. finetune_model

3. forecast

4. anomaly_detect

5. list_org_models

Connecting AI Agents

Claude Desktop

Custom Integration

Development

Security Features

License

Contributing

Support

Maintenance

Resources

Related MCP Connectors

Related MCP Servers

shopify-forecast-mcp

forecast-mcp

data-science-mcp

TimesFM-3 Studio MCP Server

1. `validate_dataset`

2. `finetune_model`

3. `forecast`

4. `anomaly_detect`

5. `list_org_models`