MarcoPolo

Here is a fact about enterprise AI that vendors would prefer you not dwell on: almost every demo works, and almost none of them reach production. The model is rarely the problem. The problem is the data - which lives in a Snowflake warehouse, and also a Postgres database, and also an S3 bucket, and also Salesforce, and also a Jira instance nobody has fully documented since 2019. Getting a model to safely read across all of that, without handing it credentials it shouldn't have or data it shouldn't keep, is unglamorous work. MarcoPolo decided to do the unglamorous work.

The company, based in Santa Clara, describes itself with a tagline that is refreshingly specific about what it is afraid of: "Scale AI across the enterprise. Without the leaks." That second sentence is the whole business. MarcoPolo sits between AI assistants - Claude, Cursor, ChatGPT, Copilot, anything that speaks the Model Context Protocol - and a company's internal systems, and it lets the AI query, correlate and act on that data through a single governed connection instead of fifty fragile ones.

"AI tools should work directly with the systems where real work happens - without friction or fragile integrations."

The pitch is deceptively plain: stop building integrations, start building with your data. Anyone who has maintained a custom connector between two systems knows the appeal. Each bespoke pipeline is technical debt with a countdown timer, one API change away from breaking at an inconvenient hour. MarcoPolo's bet is that a single secure MCP connection - scoped credentials, isolated execution, encrypted secrets, full audit logging - beats a drawer full of brittle scripts. Fewer moving parts, fewer 2am pages.

What makes this more than a slogan is the architecture underneath it. The model never gets the raw data or the raw credentials. It gets scoped access through a governed workspace, and everything it touches is logged and, if you like, piped into your SIEM. For a security team, that changes the question from "should we let AI near our data?" to "which scopes do we grant?" - a much easier conversation to have.

There is a quiet ambition here too. MarcoPolo talks about being a kind of "data computer": a place where AI can reason across systems rather than inside any single one. That framing matters, because the interesting intelligence in an enterprise usually lives in the correlation between silos - the customer record that only makes sense when you join usage logs to the CRM to the support tickets. A tool that can see across all three at once is doing something a single-source query never could.

A pivot is not a failure, though it is often mistaken for one. MarcoPolo is the second act of Immersa, a RevOps data-intelligence startup founded around 2021 that helped sales and service teams turn product-usage data into upsell and renewal signals. Immersa raised roughly $16M over its life, including a Series A with investors such as Mayfield's Navin Chaddha and the Neythri Futures Fund.

Then the Model Context Protocol arrived, and the same team saw a larger problem hiding inside the smaller one. The hard part of their old business had always been getting AI to safely reason over enterprise data. In February 2024 they made that hard part the whole company and relaunched as MarcoPolo. Sometimes the second idea was living inside the first one all along.

The unsexy layer wins. Everyone wants to build the agent. Almost nobody wants to build the secure, auditable connection between the agent and the enterprise's messy data.

The people involved are not first-timers. Co-founder and CEO Aman Singla holds a computer-science PhD from Georgia Tech and was previously CTO at Plume, with earlier engineering leadership at Qualcomm and Atheros; he co-founded a university along the way. Co-founder Aseem Chandra rounds out the leadership. The broader team is described as engineers who have built and scaled global data platforms - which, for an infrastructure company, is the relevant credential.

In 2025 the company introduced a Model Context Repository at AICamp in San Francisco, extending its work on the deceptively deep problem of delivering the right context to a language model. Context engineering is, increasingly, where enterprise AI is won or lost, and MarcoPolo has planted its flag there.

Early customers include DuploCloud, Sybill, Fresh KDS, Frore Systems, Skyflow and Theom - a mix of infrastructure, AI and hardware companies with the kind of scattered internal data that makes MarcoPolo's pitch land. It is a short list, which is what an early list looks like, and the company appends the honest phrase "and growing."