Someone on your team is already asking an AI assistant what a role should pay. They are doing it in the tool they keep open all day, and they are getting an answer that sounds confident. The question worth asking is where that answer came from.
An MCP server changes the answer. Instead of the assistant reaching into what it happens to remember from training, it requests the figure from a live source and reports what that source says. This guide explains what an MCP server is, why compensation data is a particularly bad thing to answer from memory, what the LaborIQ MCP server returns, and what it does not do.
| DEFINITION MCP SERVER MCP stands for Model Context Protocol, an open standard that lets an AI assistant request data from an outside system while a conversation is in progress. An MCP server is the piece the data owner runs. It exposes a defined set of questions the assistant is allowed to ask, and returns current values in a form the assistant can quote back to the user. |
What you will find in this guide
- What an MCP server is, in terms that matter to an HR team rather than an engineering one
- Why salary questions are the worst case for an assistant answering from memory
- What happens, step by step, when a connected assistant answers a pay question
- What the free and Pro tiers of the LaborIQ MCP server return
- Where this fits into real compensation work
- What an MCP server does not do, and what still needs a person
- How to get your team connected
- How LaborIQ supports compensation work inside AI tools
- Frequently asked questions
What is an MCP server?
Think of it as a supply line rather than a feature. An AI assistant on its own is a closed system: it can only work with what it absorbed during training and whatever you paste into the conversation. An MCP server opens a controlled door to a live system, so the assistant can fetch a current figure at the moment the question is asked.
Two things about that door matter for compensation. The first is that the data owner decides what is behind it. LaborIQ defines which questions the server answers and what each answer contains, so the assistant cannot wander into data it was never meant to see. The second is that the answer arrives attributed. The user sees where the figure came from, which is the difference between a number they can take to a hiring manager and a number they cannot.
Why salary questions are the hard case
Most questions people bring to an AI assistant tolerate an approximate answer. Compensation does not. A salary figure carries a date whether or not anyone states it, and a benchmark that was accurate eighteen months ago is not slightly stale today. It is wrong in a specific direction, and it will stay wrong in that direction across every offer built on it.
This is a known problem, and LaborIQ has written about the accuracy risks of AI-generated compensation data before there was a product answer to pair with it. The failure is not that the model is careless. It is that a model answering from memory has no way to tell you how old its memory is.
| KEY TAKEAWAY An unconnected assistant cannot distinguish between a figure it learned last year and one it learned three years ago, because both feel equally certain to it. Connecting it to a live source replaces that certainty with something you can check. |
How a connected answer works
| Step | What happens |
| You ask | A normal question in normal words, inside the assistant your team already uses. |
| The assistant recognizes the request | It identifies that the question is one the LaborIQ server can answer, and calls it rather than answering itself. |
| LaborIQ returns the data | The current benchmark for that role and market, as it stands at the moment of the request. |
| The answer is attributed | The reply carries the figure and names LaborIQ as the source, so the user knows what they are quoting. |
What the LaborIQ MCP server returns
Access comes in two tiers. The free tier is open and needs no account, which means anyone evaluating a role can get a median figure without a sales conversation first. The Pro tier is for LaborIQ customers and returns the forward-looking fields that make a benchmark usable for planning rather than only for checking.
| Field | Tier | What it tells you |
| Median salary | Free | What the role pays today in the market you named. |
| Salary forecast | Pro | Where the figure is heading, which is what a range built now has to survive. |
| Labor supply index | Pro | How much competition you face for the role in that market. |
| Year-over-year wage growth | Pro | How fast this specific role has moved, rather than how fast wages moved in general. |
| Full job catalog | Pro | Coverage beyond the defined free list, across the LaborIQ job library. |
When a free user asks about a role outside the open list, the answer says so and points to the Pro tier rather than guessing. That is deliberate. A server that improvises when it runs out of data would undo the reason to connect it in the first place.
Where this fits into real compensation work
| Situation | What connected data changes |
| Building an offer | The recruiter checks the current median while drafting, instead of reusing the range from the last hire in that job family. |
| Answering a manager mid-conversation | The question gets a sourced answer in the same thread, rather than becoming a task for later. |
| Opening a role in a new market | Geographic differences surface before the requisition is approved, not after two failed searches. |
| Sanity-checking a band | A quick comparison against the market before a full analysis is commissioned. |
| Planning headcount | Forecast and supply fields turn a budget conversation into a market conversation. |
What an MCP server does not do
It returns a market figure. It does not decide what you should pay. The gap between those two things is the entire job of a compensation function, and connecting a data source does not close it.
| CRITICAL DISTINCTION A median for a job title is not a match for your job. The market figure assumes a scope, a level and a set of responsibilities that may not be yours. Treating a returned number as a finished answer reproduces the oldest mistake in compensation work, only faster. The server gives you the market. Matching your role to it is still yours to do. |
The same caution applies to the questions around the number. Internal equity, your pay philosophy, budget reality and the disclosure rules in the states where you operate are all outside what any data feed can answer. A connected assistant makes the market half faster. It leaves the other half where it was.
Getting your team connected
Connecting an MCP server is a configuration step in the AI tool itself, not an installation on anyone’s machine. The pattern is the same across assistants that support the protocol.
| Step | What it involves |
| Open your assistant’s connector settings | The section where external tools and data sources are added. |
| Add the LaborIQ server | Point it at the LaborIQ MCP endpoint. |
| Authenticate for Pro access | Free access needs no account. Pro fields require your LaborIQ credentials. |
| Ask a question | No new interface to learn. The assistant now reaches for LaborIQ when a pay question comes up. |
| IMPORTANT Compensation is among the most access-controlled data any HR team handles, so bring IT and legal into this before a wide rollout rather than after. Agree who is allowed Pro access, which assistants are approved, and what employees may paste into a conversation. The connection is straightforward. The governance around it is the part worth planning. |
How LaborIQ supports compensation work inside AI tools
LaborIQ maintains salary benchmarks for over 20,000 unique job titles across U.S. markets, built from 18 trillion data points and validated against 8.6 million real company pay stubs, spanning 1,600 industries. The MCP server is one way to reach that data. Salary Answers is the other, for the work that belongs in a dedicated tool rather than in a chat.
| LABORIQ PLATFORM Compensation Intelligence Built for HR Leaders ✓ Salary benchmarks at the 25th, 50th, 75th and 90th percentiles for every U.S. role ✓ Role matching across thousands of job families and levels ✓ Geographic pay differential analysis for remote and multi-location teams ✓ Salary forecasts and labor supply signals for planning ahead ✓ Live access from the AI tools your team already works in → Request a Free Demo at laboriq.co/request-demo |
Frequently asked questions
What does MCP stand for?
Model Context Protocol. It is an open standard that lets an AI assistant request data from an outside system during a conversation, rather than answering only from what it learned in training.
Do I need a LaborIQ account to use the MCP server?
Not for the free tier, which returns median salary for a defined list of roles without an account. The Pro fields, including salary forecast, labor supply index and year-over-year wage growth, are for LaborIQ customers and require your LaborIQ credentials.
Which AI tools can connect to it?
Any assistant that supports the Model Context Protocol. Because MCP is an open standard rather than a product integration, support is decided by the assistant, and the list grows as more tools adopt it.
Is this the same as an API?
They serve different users. An API is built for developers writing software against it. An MCP server is built for an AI assistant to call on behalf of a person asking a question in plain language, with no code involved.
Does the assistant see our internal salary data?
No. The server sends market data outward. It does not read your systems. Anything about your own employees enters the conversation only if someone types or pastes it, which is exactly why an internal policy on what may be shared is worth agreeing before a wide rollout.
How current is the data the server returns?
The server returns the benchmark as it stands in LaborIQ at the moment of the request, which is the point of connecting it. That is different from an assistant answering from training data, where the age of the figure is unknown to both of you.
Can we rely on the returned figure to set pay?
Use it as the market input, not as the decision. A returned median assumes a scope and level that may not match your role, and it says nothing about your internal equity, your pay philosophy or your budget. Match the role first, then decide.
