Glean MCP Server Cost: What FlexCredits Actually Bill

The question comes up the week a platform team wires Cursor or Claude Code into company knowledge. Somebody finds the Glean MCP server, the setup takes an afternoon, engineers love it, and then finance asks what it costs. At that point most teams search for the answer and find a confident, wrong one.

The wrong answer is that it is free. It is repeated because there genuinely is an MIT-licensed Glean MCP package on GitHub, and because Glean's pricing page publishes nothing at all, so anyone looking for a number finds the license file instead. This article sets out what Glean actually documents about MCP billing, which of its APIs are metered and which are not, and what to put in front of your account team before the usage arrives on an invoice.

Does the Glean MCP server cost extra?

Yes. The About Glean MCP server page carries a Pricing section, and it reads in full: "This feature is subject to usage-based pricing and these supplemental FlexCredit terms." There is no allowance mentioned, no bundled tier, and no exemption for MCP traffic.

The Enterprise Flex documentation says the same thing from the other direction, and more usefully, because it names the surface. Under Developer Tools, Glean lists "Client APIs" covering the Search API, Chat API and Agents API, with the usage model "Consumes FlexCredits at current rate card" and the access surfaces given as "Client API endpoint, MCP Server, Agent toolkits". The MCP server is not a separate product with its own billing. It is a front door onto the Client APIs, and those are metered.

That is also why the MIT license is a red herring. The package on GitHub is client code that speaks the Model Context Protocol on one side and calls Glean's API on the other. You are free to read it, fork it and ship it. What it calls still bills. Glean's developer guide is careful to separate the pieces: the Remote MCP Server is "tenant-specific and permission-aware", each person uses their own Glean access, and it is explicitly "different from the public Docs MCP Server, which gives a coding assistant access to Glean developer documentation". The public docs server is genuinely free and genuinely useless for company knowledge, because it only serves Glean's own documentation.

Which Glean APIs are free and which consume FlexCredits?

The split is clean once you see it: putting data into Glean is unlimited, getting answers out of it is metered. Glean's Enterprise Flex table divides Developer Tools into three rows, and only one of them draws credits.

This is the sentence worth carrying into a budget meeting: you can index a hundred million documents, build every custom connector you want and authenticate every user for nothing, and then pay per question. It is a defensible model, and it is the opposite of the per-seat assumption most buyers arrive with. It also means MCP traffic is uniquely expensive in one specific way, because an AI coding host does not ask one question. It calls tools in a loop, several times per prompt, and each of those is a Client API call.

What is included in a Glean Enterprise Flex seat?

Fast Mode queries are unlimited. Thinking Mode and Adaptive Reasoning Mode queries with standard models are included up to 100 per user per week. Everything else on the list consumes FlexCredits.

The footnotes underneath that table matter more than the table, and they are where the 100 per week allowance stops meaning what it appears to mean. Three of them, verbatim:

• "Thinking Queries that consume > 30 FlexCredits will be counted as multiple queries."
• "One prompt and response forms a single query, asking follow-up queries are counted as additional queries."
• "For Assistant queries that route to an agent, the agent run will consume FlexCredits."

Read together, the weekly allowance is not 100 questions. It is 100 units, where a heavy question counts as several, every follow-up in a conversation counts again, and a question the Assistant decides to route to an agent leaves the allowance altogether and bills as an agent run. None of that is unreasonable metering. It is simply not what a buyer hears when a seller says "100 thinking queries a week are included", and it is the single most common reason a Glean bill lands above forecast.

How much is a FlexCredit in dollars?

Glean does not publish it. Every metered line in the documentation says "at current rate card", and the rate card itself is not on a public page. That is a real limit on what anyone can honestly tell you, including us. If you read an article that gives you a dollar figure per FlexCredit and cites no contract, it is inferred.

What does exist publicly is third-party procurement data, and it should be labelled as exactly that every time. Vendr reports an average and median Glean contract of $98,890 a year across a range of $29,880 to $208,897. SpendHound reports $142,019 a year as an SMB average and $259,933 for enterprise. Those are aggregates of what buyers paid, not a price list, and they cannot be decomposed into a credit rate.

So the practical move is to make the rate card a deliverable of the negotiation rather than a discovery after signature. Three things are worth asking for in writing: the FlexCredit rate card itself with the premium model list attached, the credit consumption range for a Client API call made over MCP rather than for Assistant queries, and a worked example at your own expected volume. The third one is the test. A vendor that cannot model your usage from their own rate card is telling you something.

One documented lever is worth raising in the same conversation. Glean states that "Enterprise Flex Seats and FlexCredits are discounted for customers who choose to supply their own LLM keys and/or self-host Glean in their private cloud instance to compensate for compute and model infrastructure costs the customer bears." If you already hold model capacity with a provider, that is a discount you can ask for by name.

How do I track and cap Glean MCP spend?

Through the Usage dashboard under Admin console then Usage, which requires the Super Admin, Admin or Billing Moderator role. It shows FlexCredits used against total available, with trends by day or week, and a Breakdown section organised into four tabs: by product types, by users, by models and by departments.

The product types tab is the one that answers the MCP question directly. It splits usage into Assistant, Agents and API, and Glean describes the API view as showing "a table that lists specific APIs and their FlexCredits consumed, such as Search API". Since MCP traffic runs through the Client APIs, that is where it surfaces. The models view flags premium models with a "$ Premium" tag, which is useful because premium models have no included allowance at all.

Two limits on the dashboard are worth knowing before you rely on it. The per-user table is only available "for any single-month selection from July 2026 onward" and not when you select All time. And organisations on a legacy Glean Enterprise plan do not get the dashboard at all: Glean states that Admin console then Usage "shows a notice that this dashboard is no longer available instead of the dashboard itself, because that dashboard measures Enterprise Flex pricing". If you are on an older agreement and were planning to govern MCP rollout with usage data, check which plan you are on first.

Controls do exist once you can see the numbers. Glean supports usage limits and alerts at the organisation level, spending limits for specific users, and limits within a department, and the department table exports to CSV for chargeback and showback. For a rollout to engineers, department limits are the sane default, because a coding host in a loop is the workload most likely to surprise you.

One operational note that has nothing to do with billing and everything to do with whether the rollout survives contact with your engineers. The remote MCP server becomes a live dependency inside the inner development loop, so when the endpoint is unreachable every session silently loses company context rather than failing loudly. It belongs on the same uptime monitoring you run for internal APIs, not in the category of things somebody notices eventually.

What to do with this before you sign

If Glean is already deployed and the question is whether to turn on MCP, the answer is usually yes, with department-level limits set first and the usage dashboard checked weekly for the first month. The capability is real, it is permission-aware, each person uses their own Glean access, and the whole point of an index is that other tools can query it.

If the question is whether to buy Glean in order to get MCP, price the whole thing rather than the feature. A seat that meters the retrieval your developers will do all day is a different purchase from a seat that does not, and the rate card is the document that decides which one you are buying. Our full breakdown of what Glean publishes and what it does not is on Glean pricing, and the shortlist buyers usually build alongside it is on Glean alternatives.

It is also worth checking whether you need a third-party index at all for the specific job in front of you. If the requirement is company knowledge inside an AI assistant your team already pays for, Claude Ask Your Org does it with no index and no separate credit meter, at the cost of coverage that depends on which individuals authenticated which connectors. If the requirement is one complete, cited answer across every company system, connected centrally and priced per seat with no second meter to forecast, that is what enterprise search software is for, and you can put a real question into the search box at the top of this page and see what comes back.