The Mannequin Context Protocol (MCP) is genuinely helpful. It provides individuals who develop AI instruments a standardized approach to name features and entry information from exterior techniques. As an alternative of constructing customized integrations for every information supply, you possibly can expose databases, APIs, and inner instruments by a standard protocol that any AI can perceive.
Nonetheless, I’ve been watching groups undertake MCP over the previous yr, and I’m seeing a disturbing sample. Builders are utilizing MCP to shortly join their AI assistants to each information supply they’ll discover—buyer databases, assist tickets, inner APIs, doc shops—and dumping all of it into the AI’s context. And since the AI is wise sufficient to type by a large blob of knowledge and pick the elements which can be related, all of it simply works! Which, counterintuitively, is definitely an issue. The AI cheerfully processes large quantities of knowledge and produces cheap solutions, so no person even thinks to query the method.
That is information hoarding. And like bodily hoarders who can’t throw something away till their properties grow to be so cluttered they’re unliveable, information hoarding has the potential to trigger severe issues for our groups. Builders study they’ll fetch much more information than the AI wants and supply it with little planning or construction, and the AI is wise sufficient to take care of it and nonetheless give good outcomes.
When connecting a brand new information supply takes hours as a substitute of days, many builders don’t take the time to ask what information truly belongs within the context. That’s how you find yourself with techniques which can be costly to run and inconceivable to debug, whereas a complete cohort of builders misses the possibility to study the essential information structure expertise they should construct strong and maintainable functions.
How Groups Be taught to Hoard
Anthropic launched MCP in late 2024 to offer builders a common approach to join AI assistants to their information. As an alternative of sustaining separate code for connectors to let AI entry information from, say, S3, OneDrive, Jira, ServiceNow, and your inner DBs and APIs, you utilize the identical easy protocol to supply the AI with all kinds of knowledge to incorporate in its context. It shortly gained traction. Firms like Block and Apollo adopted it, and groups in every single place began utilizing it. The promise is actual; in lots of instances, the work of connecting information sources to AI brokers that used to take weeks can now take minutes. However that velocity can come at a price.
Let’s begin with an instance: a small staff engaged on an AI device that reads buyer assist tickets, categorizes them by urgency, suggests responses, and routes them to the proper division. They wanted to get one thing working shortly however confronted a problem: That they had buyer information unfold throughout a number of techniques. After spending a morning arguing about what information to tug, which fields have been vital, and find out how to construction the combination, one developer determined to simply construct it, making a single getCustomerData(customerId) MCP device that pulls all the pieces they’d mentioned—40 fields from three completely different techniques—into one massive response object. To the staff’s aid, it labored! The AI fortunately consumed all 40 fields and began answering questions, and no extra discussions or selections have been wanted. The AI dealt with all the brand new information simply high-quality, and everybody felt just like the venture was heading in the right direction.
Day two, somebody added order historical past so the assistant might clarify refunds. Quickly the device pulled Zendesk standing, CRM standing, eligibility flags that contradicted one another, three completely different identify fields, 4 timestamps for “final seen,” plus whole dialog threads, and mixed all of them into an ever-growing information object.
The assistant saved producing reasonable-looking solutions, at the same time as the information it ingested saved rising in scale. Nonetheless, the mannequin now needed to wade by hundreds of irrelevant tokens earlier than answering easy questions like “Is that this buyer eligible for a refund?” The staff ended up with an information structure that buried the sign in noise. That extra load put stress on the AI to dig out that sign, resulting in severe potential long-term issues. However they didn’t understand it but, as a result of the AI saved producing reasonable-looking solutions. As they added extra information sources over the next weeks, the AI began taking longer to reply. Hallucinations crept in that they couldn’t monitor all the way down to any particular information supply. What had been a very worthwhile device turned a bear to keep up.
The staff had fallen into the information hoarding lure: Their early fast wins created a tradition the place individuals simply threw no matter they wanted into the context, and ultimately it grew right into a upkeep nightmare that solely received worse as they added extra information sources.
The Abilities That By no means Develop
There are as many opinions on information structure as there are builders, and there are normally some ways to unravel anybody downside. One factor that just about everybody agrees on is that it takes cautious selections and many expertise. But it surely’s additionally the topic of a number of debate, particularly inside groups, exactly as a result of there are such a lot of methods to design how your utility shops, transmits, encodes, and makes use of information.
Most of us fall into just-in-case pondering at one time or one other, particularly early in our careers—pulling all the information we would presumably want simply in case we’d like it somewhat than fetching solely what we’d like once we really need it (which is an instance of the alternative, just-in-time pondering). Usually once we’re designing our information structure, we’re coping with speedy constraints: ease of entry, dimension, indexing, efficiency, community latency, and reminiscence utilization. However once we use MCP to supply information to an AI, we are able to typically sidestep lots of these trade-offs…briefly.
The extra we work with information, the higher we get at designing how our apps use it. The extra early-career builders are uncovered to it, the extra they study by expertise why, for instance, System A ought to personal buyer standing whereas System B owns cost historical past. Wholesome debate is a crucial a part of this studying course of. By means of all of those experiences, we develop an instinct for what “an excessive amount of information” seems to be like—and find out how to deal with all of these difficult however essential trade-offs that create friction all through our initiatives.
MCP can take away the friction that comes from these trade-offs by letting us keep away from having to make these selections in any respect. If a developer can wire up all the pieces in just some minutes, there’s no want for dialogue or debate about what’s truly wanted. The AI appears to deal with no matter information you throw at it, so the code ships with out anybody questioning the design.
With out all of that have making, discussing, and debating information design selections, builders miss the possibility to construct essential psychological fashions about information possession, system boundaries, and the price of shifting pointless information round. They spend their adolescence connecting as a substitute of architecting. That is one other instance of what I name the cognitive shortcut paradox—AI instruments that make growth simpler can forestall builders from constructing the very expertise they should use these instruments successfully. Builders who rely solely on MCP to deal with messy information by no means study to acknowledge when information structure is problematic, identical to builders who rely solely on instruments like Copilot or Claude Code to generate code by no means study to debug what it creates.
The Hidden Prices of Information Hoarding
Groups use MCP as a result of it really works. Many groups rigorously plan their MCP information structure, and even groups that do fall into the information hoarding lure nonetheless ship profitable merchandise. However MCP continues to be comparatively new, and the hidden prices of knowledge hoarding take time to floor.
Groups typically don’t uncover the issues with an information hoarding method till they should scale their functions. That bloated context that hardly registered as a price on your first hundred queries begins exhibiting up as an actual line merchandise in your cloud invoice once you’re dealing with tens of millions of requests. Each pointless discipline you’re passing to the AI provides up, and also you’re paying for all that redundant information on each single AI name.
Any developer who’s handled tightly coupled courses is aware of that when one thing goes flawed—and it all the time does, ultimately—it’s rather a lot more durable to debug. You typically find yourself coping with shotgun surgical procedure, that actually disagreeable state of affairs the place fixing one small downside requires adjustments that cascade throughout a number of elements of your codebase. Hoarded information creates the identical type of technical debt in your AI techniques: When the AI provides a flawed reply, monitoring down which discipline it used or why it trusted one system over one other is troublesome, typically inconceivable.
There’s additionally a safety dimension to information hoarding that groups typically miss. Each piece of knowledge you expose by an MCP device is a possible vulnerability. If an attacker finds an unprotected endpoint, they’ll pull all the pieces that device offers. In the event you’re hoarding information, that’s your whole buyer database as a substitute of simply the three fields truly wanted for the duty. Groups that fall into the information hoarding lure discover themselves violating the precept of least privilege: Purposes ought to have entry to the information they want, however no extra. That may carry an unlimited safety danger to their complete group.
In an excessive case of knowledge hoarding infecting a complete firm, you would possibly uncover that each staff in your group is constructing their very own blob. Help has one model of buyer information, gross sales has one other, product has a 3rd. The identical buyer seems to be utterly completely different relying on which AI assistant you ask. New groups come alongside, see what seems to be working, and replica the sample. Now you’ve received information hoarding as organizational tradition.
Every staff thought they have been being pragmatic, delivery quick, and avoiding pointless arguments about information structure. However the hoarding sample spreads by a corporation the identical means technical debt spreads by a codebase. It begins small and manageable. Earlier than you understand it, it’s in every single place.
Sensible Instruments for Avoiding the Information Hoarding Entice
It may be actually troublesome to teach a staff away from information hoarding once they’ve by no means skilled the issues it causes. Builders are very sensible—they wish to see proof of issues and aren’t going to sit down by summary discussions about information possession and system boundaries when all the pieces they’ve completed to date has labored simply high-quality.
In Studying Agile, Jennifer Greene and I wrote about how groups resist change as a result of they know that what they’re doing at this time works. To the particular person attempting to get builders to alter, it could seem to be irrational resistance, nevertheless it’s truly fairly rational to push again in opposition to somebody from the surface telling them to throw out what works at this time for one thing unproven. However identical to builders ultimately study that taking time for refactoring speeds them up in the long term, groups must study the identical lesson about deliberate information design of their MCP instruments.
Listed below are some practices that may make these discussions simpler, by beginning with constraints that even skeptical builders can see the worth in:
- Construct instruments round verbs, not nouns. Create
checkEligibility()orgetRecentTickets()as a substitute ofgetCustomer(). Verbs drive you to consider particular actions and naturally restrict scope. - Discuss minimizing information wants. Earlier than anybody creates an MCP device, have a dialogue about what the smallest piece of knowledge they should present for the AI to do its job is and what experiments they’ll run to determine what the AI really wants.
- Break reads other than reasoning. Separate information fetching from decision-making once you design your MCP instruments. A easy
findCustomerId()device that returns simply an ID makes use of minimal tokens—and won’t even should be an MCP device in any respect, if a easy API name will do. ThengetCustomerDetailsForRefund(id)pulls solely the precise fields wanted for that call. This sample retains context centered and makes it apparent when somebody’s attempting to fetch all the pieces. - Dashboard the waste. The very best argument in opposition to information hoarding is exhibiting the waste. Observe the ratio of tokens fetched versus tokens used and show them in an “info radiator” fashion dashboard that everybody can see. When a device pulls 5,000 tokens however the AI solely references 200 in its reply, everybody can see the issue. As soon as builders see they’re paying for tokens they by no means use, they get very keen on fixing it.
Fast odor check for information hoarding
- Software names are nouns
(getCustomer())as a substitute of verbs(checkEligibility()). - No one’s ever requested, “Do we actually want all these fields?”
- You’ll be able to’t inform which system owns which piece of knowledge.
- Debugging requires detective work throughout a number of information sources.
- Your staff hardly ever or by no means discusses the information design of MCP instruments earlier than constructing them.
Trying Ahead
MCP is a straightforward however highly effective device with monumental potential for groups. However as a result of it may be a critically essential pillar of your whole utility structure, issues you introduce on the MCP stage ripple all through your venture. Small errors have enormous penalties down the highway.
The very simplicity of MCP encourages information hoarding. It’s a straightforward lure to fall into, even for knowledgeable builders. However what worries me most is that builders studying with these instruments proper now would possibly by no means study why information hoarding is an issue, and so they gained’t develop the architectural judgment that comes from having to make laborious selections about information boundaries. Our job, particularly as leaders and senior engineers, is to assist everybody keep away from the information hoarding lure.
While you deal with MCP selections with the identical care you give any core interface—protecting context lean, setting boundaries, revisiting them as you study—MCP stays what it needs to be: a easy, dependable bridge between your AI and the techniques that energy it.