# MarcoPolo

> MarcoPolo is production-ready MCP middleware that connects enterprise data to any LLM. It gives agentic AI a secure, persistent workspace to query, correlate, and analyze data across 50+ real systems — without pipelines, exports, or code.

MarcoPolo is built by Immersa, Inc., headquartered in Santa Clara, CA. The product is available at https://marcopolo.dev and the MCP server endpoint is https://mcp.marcopolo.dev.

## What is MarcoPolo?

MarcoPolo is an AI data runtime — a layer between your LLM and your enterprise data systems. Unlike a simple API connector, MarcoPolo provides a full execution environment: DuckDB for SQL queries, Python for transforms, and shell for everything else. Results persist between steps so your AI can iterate on live data, not start over each time.

MarcoPolo implements the Model Context Protocol (MCP), making it compatible with Claude, ChatGPT, Cursor, GitHub Copilot, VS Code, Replit, and any MCP-compatible client.

## Who is MarcoPolo for?

- **AI developers** building agentic workflows that need real data access
- **Enterprise teams** that want their AI to query internal systems securely
- **Data and engineering teams** replacing manual exports with AI-driven analysis
- **CTOs and AI leaders** who need governed, auditable AI data access

## Key Capabilities

- **50+ data source connectors**: Cloud data warehouses (Snowflake, BigQuery, Redshift, Databricks, Azure Synapse), relational databases (PostgreSQL, MySQL, SQL Server, Oracle, MongoDB), big data platforms (Athena, Presto, Hive, ClickHouse, Elasticsearch), time series and NoSQL (InfluxDB, Prometheus, Cassandra, DynamoDB), and SaaS tools (Salesforce, HubSpot, Intercom, Google Analytics, JIRA)
- **Cross-source joins**: Combine data from Salesforce, Postgres, and S3 in a single AI conversation
- **Isolated execution**: Every AI session runs inside an isolated Kubernetes container with scoped credentials and zero data leakage to the LLM
- **Persistent workspace**: DuckDB, Python, and shell available; intermediate results survive between steps
- **Schema awareness**: MarcoPolo learns your data landscape — tables, columns, relationships, documentation, and example queries are loaded automatically
- **Multi-LLM support**: Works with Claude (Web, Desktop, Code), ChatGPT, Cursor, VS Code Copilot, Copilot Studio, Replit, and any streamable HTTP MCP client
- **Enterprise SSO**: One-click auth with Google, Microsoft, GitHub, or Enterprise SSO

## Security Model

MarcoPolo's security is built around three principles:
1. **Scoped credentials** — the AI gets access only to what the task requires
2. **Isolated containers** — every session is a clean, sandboxed Kubernetes environment; no state leaks between sessions
3. **Zero leakage** — no data or credentials are passed to the LLM or leave your environment

MarcoPolo is SOC 2 Type II compliant. Security details are available at https://marcopolo.dev/trust.

## How MarcoPolo Works (3 Steps)

1. **Connect MarcoPolo to your AI client** — Add MarcoPolo as an MCP server in Claude, ChatGPT, Cursor, or any MCP client. One-click auth with Google, Microsoft, GitHub, or Enterprise SSO.
2. **Authorize your data sources** — Select which systems your AI can access. Set scoped permissions through the governed interface.
3. **Let AI do the real work** — Your AI builds and executes queries inside a secure container, iterates with live data, and returns actionable results.

## How MarcoPolo differs from alternatives

- **vs. vector databases**: MarcoPolo queries live, structured data across real systems. It does not require pre-indexing or embedding. Results reflect the current state of your data.
- **vs. direct LLM tool calls / function calling**: MarcoPolo provides a persistent execution environment with real compute (DuckDB SQL, Python). The AI can iterate across multiple steps without context limits.
- **vs. ETL pipelines**: No pipelines, no scheduled exports, no stale data. Queries run on demand against live sources.
- **vs. other MCP servers**: MarcoPolo is a runtime, not just a connector. It handles cross-source joins, session persistence, schema context, and enterprise security in one governed layer.

## Pricing

MarcoPolo offers a free tier to get started with MCP. Enterprise pricing is available for teams needing VPC deployment, advanced RBAC, and SLA guarantees. Details at https://marcopolo.dev/enterprise#pricing.

## Frequently Asked Questions

**What is MarcoPolo?**
MarcoPolo is production-ready MCP middleware that gives enterprise AI a secure workspace to query live data across 50+ systems — including databases, data warehouses, and SaaS tools — without any pipelines or exports.

**What is MCP middleware?**
MCP (Model Context Protocol) middleware sits between an AI client (like Claude or ChatGPT) and your data systems, handling authentication, query execution, security, and context management. MarcoPolo is an MCP middleware layer that adds a full execution environment on top of raw MCP connectivity.

**Which LLMs does MarcoPolo support?**
MarcoPolo works with Claude (Web, Desktop, and Code), ChatGPT, Cursor, GitHub Copilot, VS Code, Copilot Studio, Replit, and any MCP-compatible client that supports streamable HTTP with OAuth.

**What data sources can MarcoPolo connect to?**
MarcoPolo supports 50+ sources including Snowflake, BigQuery, Redshift, Databricks, PostgreSQL, MySQL, SQL Server, Oracle, MongoDB, Salesforce, HubSpot, JIRA, Google Analytics, Amazon Athena, ClickHouse, Elasticsearch, InfluxDB, and more. New connectors are added weekly.

**How does MarcoPolo handle security and access control?**
Every AI session runs in an isolated Kubernetes container with scoped credentials. No data or credentials are passed to the LLM. MarcoPolo supports RBAC, enterprise SSO, and is SOC 2 Type II certified.

**Does MarcoPolo support VPC deployment?**
Yes, VPC deployment is available on the Enterprise plan. Contact the team at https://marcopolo.dev/contact.

**Is there a free tier?**
Yes. You can start for free at https://marcopolo.dev/#connect using the MCP server at https://mcp.marcopolo.dev.

**How is MarcoPolo different from a vector database?**
MarcoPolo queries live structured data directly — it does not require pre-indexing, embeddings, or chunking. It is designed for operational and analytical queries against real-time enterprise systems, not semantic search over documents.

**Where is data processed?**
Data is processed inside isolated containers within your governed environment. Nothing is passed to the LLM provider — only the results your AI needs to continue the conversation.

## Key Pages

- Homepage: https://marcopolo.dev
- Enterprise & Pricing: https://marcopolo.dev/enterprise
- Security & Trust: https://marcopolo.dev/trust
- Documentation: https://docs.marcopolo.dev
- Blog: https://marcopolo.dev/blog
- Contact: https://marcopolo.dev/contact
- MCP Server URL: https://mcp.marcopolo.dev

## Company

MarcoPolo is a product of Immersa, Inc.
Address: 4677 Old Ironsides Dr, #315, Santa Clara, CA 95054
LinkedIn: https://www.linkedin.com/company/marcopolodev/
Slack Community: https://join.slack.com/t/marcopolo-dev/shared_invite/zt-3svcueyac-l~ugqJBrjkwWI6VYVZ005A