By When an AI agent needs to answer a question that spans Salesforce, HubSpot, and Zendesk, an MCP gateway fires a separate tool call to each source. Every extra call adds latency, burns context tokens, and increases the probability of reconciliation errors. As the context window fills with raw payloads, context rot sets in and the agent’s reasoning quality drops.
In this interview on TFiR, Jean Lafleur, Co-founder and Chief Operating Officer at Airbyte, walks through why the MCP gateway round-trip model breaks down at scale and how prematerializing context with a dedicated context store reduces multi-source lookups to a single tool call.
Guest: Jean Lafleur, Co-founder and Chief Operating Officer at Airbyte
Show: TFiR
Here is what every platform engineer and AI agent developer needs to know.
Technical Deep Dive
Q: Why does an MCP gateway generate multiple tool calls when an AI agent queries several data sources?
Jean Lafleur, Co-founder and Chief Operating Officer at Airbyte, explains that an MCP gateway routes each source lookup as a separate round trip. If an agent asks about the latest status of an account that lives in Salesforce, HubSpot, and Zendesk, the gateway issues at least one tool call per source. All of those results must then be reconciled at runtime, which multiplies latency, burns context tokens, and introduces compounding reconciliation errors.
“You will have to have many tool calls for each source. Then you reconcile all of that during runtime. And so that’s very long, very costly in terms of tokens, context token.”
Jean Lafleur, Co-founder and Chief Operating Officer, Airbyte
Q: What is context rot and how does it degrade AI agent accuracy?
Context rot occurs when raw source payloads flood the context window during a multi-source lookup. As the window fills, the agent has less room to reason accurately. Lafleur notes that the more bloated the context window, the less reliable the agent’s outputs become, making context rot a compounding problem for agents handling complex, multi-step tasks.
“The raw payloads really flood the context window. And so at that point your agent becomes well, less and less smart because the more context window you have, you get context rot in there.”
Jean Lafleur, Co-founder and Chief Operating Officer, Airbyte
Q: How does prematerializing context reduce tool calls for AI agents?
Airbyte’s Context Store prematerializes the context before any agent query arrives. Entity resolution happens ahead of runtime so that when a query comes in, the context store already knows which entity maps to which records across all sources. The result is a single tool call that returns a complete, reconciled answer rather than many partial calls that must be stitched together at query time.
“We prematerialize the context. We do all entity resolution. And so you have a request, you just let that request go to the context store. We know which entity, we have all the information and it’s just one tool call at that point to give you a complete comprehensive answer.”
Jean Lafleur, Co-founder and Chief Operating Officer, Airbyte
Q: Can Airbyte’s Context Store work on top of an existing data warehouse?
Yes. Lafleur describes the Context Store as flexible: it can operate as a standalone component or sit on top of an existing warehouse. This means teams that have already invested in a warehouse do not need to replace it to get the single-call benefit for their agents.
“Our Context Store can live by itself or live on top of your warehouse.”
Jean Lafleur, Co-founder and Chief Operating Officer, Airbyte
Q: How does data sovereignty fit into an agent pipeline built on Airbyte?
Lafleur traces sovereignty back to Airbyte’s open source roots. Because the project began as a self-hosted, open source standard, users can keep sensitive data entirely inside their own environment. Cloud-hosted deployment is available for data that does not require that restriction. The Flex offer, which emerged from that history, formalizes the choice between cloud and self-hosted paths so teams can match deployment model to their governance requirements.
“If it’s something sensitive and in your environment you can have access to it.”
Jean Lafleur, Co-founder and Chief Operating Officer, Airbyte
Q: How large is the Airbyte connector ecosystem and what does that mean for agent data access?
Airbyte ships more than 600 connectors and its community has built more than 35,000 custom connectors for specific use cases. That breadth of coverage means an agent context store fed by Airbyte can draw from a very wide range of sources, whether they are standard SaaS platforms or niche internal systems, without requiring custom integration work for each one.
“We’ve got more than 600 and actually our users have built more than 35,000 custom connectors for their own needs.”
Jean Lafleur, Co-founder and Chief Operating Officer, Airbyte
Resources & Documentation
- Airbyte, open source data integration platform with 600+ connectors and a context store for AI agents
- Airbyte on GitHub, source repository for the open source project and community-built connectors
***
👇 Click to Read Full Raw Transcript
Swapnil Bhartiya: Is it possible for you to walk our audience? Because you know, the fact is that stating one thing is easy, but actually walking through that, this is how it works, this is how we ensure. I mean, you talked about sovereignty, but let’s also talk about giving access to data and also what data they can access and then what they can do with the data. The governance model as well, I think
Jean Lafleur: it goes back to the history of Airbyte. We started as an open source project, we became the open source standard very, very fast. And open source is self hosted. And with the open source we managed to, you know, expand the number of connectors we were building. We’ve got more than 600 and actually our users have built more than 35,000 custom connectors for their own needs. And that is locally or also cloud hosted. And so being open source and safos, that’s how we developed that Flex offer. And that point, our goal is that wherever your data is, if it’s cloud hosted and it’s okay for you to be cloud hosted, go for it. Or if it’s something sensitive and in your environment you can have access to it. And with the agents, the difference is you have heard, I’m sure about MCP gateway where agents have a access to the tools through an MCP gateway. And the issue with the MCP gateway is that for every source you have at least a round trip. So let’s say you’re asking what is the latest with that account? And that account lives in Salesforce, Hotspot, Zendesk, in many tools. At that point you will have to have many tool calls for each source. Then you reconcile all of that during runtime. And so, and that’s very long, very costly in terms of tokens, context token. And you also, it’s usually plausibly wrong in the sense that by doing all of that you’re prone to making errors. In addition to that, you know, the raw payloads really flood the wind, the context window. And so at that point your, your agent becomes, well, less and less smart because the more context window you, you have context rot in, in there. And what we solve with agents is that we prematerialize the context. So with our context store that can live by itself or live on top of your warehouse. At that point we do all entity resolution. And so you have a request, you just that request go to the context store. We know which entity, we have all the information and it’s just one tool call at that point to give you a complete comprehensive answer. And at that point it’s faster, a lot less context tokens used. So that’s pretty important for agents when you have multiple tasks.
