How much to charge for an MCP server
How much to charge for an MCP server comes down to what the client's people can do once their AI tools can reach the system, and how safely they need to do it. A server that lets a sales team ask their AI for any customer's order history saves hours of digging, so price it on those hours. One that lets support staff update tickets from a chat changes how a whole team works, so it sits higher. Then price the parts that drive effort: how many tools, read or write access, and local or remote.
Claude Code makes the code itself quick: the tool definitions, the calls to the client's API, the error handling. What takes time is everything around it: getting API access to the client's system, deciding exactly what each tool may read or change, setting up sign-in for a shared server, and testing that the AI picks the right tool. Build the quote around those.
Find the job before you quote
What an MCP server does for a client and what to price against
| Job | What changes for the team | Price the build against |
|---|---|---|
| Look things up | Staff ask their AI for records instead of searching the system by hand | Hours of lookups per week, across the team |
| Pull reports | Numbers that took a manual export are a question away | Time spent on recurring reports |
| Update records | Staff create or change tickets, orders or entries from a chat | Admin time saved, and errors from re-typing avoided |
| Connect an internal tool | A system with no public integration becomes usable by AI | What the client would pay for a custom integration instead |
| Ship a product feature | The client's own customers connect their AI tools to the client's product | New customers or retention the feature supports |
Ask for the client's own numbers: how many people would use it, which questions they answer by hand today, and how long each takes. If they cannot say, measure it in a short paid discovery. Paid discovery for AI projects covers how to sell it. Many clients are still unsure what an MCP server is, so what MCP servers are and why they matter is a good link to send before the call.
The CCC MCP Server Quote
Six parts of an MCP server quote
| Part | What it prices | Question to ask the client |
|---|---|---|
| Access and discovery | Getting API access, test accounts and docs for the client's system | Who can give us an API key or test account, and is there a sandbox? |
| Tools | Each action the AI can take, its inputs and its output | What are the five questions or tasks people want to hand to their AI first? |
| Read or write | Whether tools only fetch data or can also change it, plus the safety checks | Should the AI ever create, edit or delete anything on its own? |
| Transport and hosting | Local server on each machine, or one remote server the team connects to | Will one person use this, or the whole team? |
| Sign-in and permissions | Shared key, or each user signing in with their own account and rights | Should everyone see the same data, or only what their own login allows? |
| Keep plan | Fixes, API changes, new tools, monitoring | Who reports problems, and how fast do they need fixing? |
Tools and read or write. Price these as a fraction of one year of the value from the table above, so the client can see the server pay for itself. A handful of read-only lookup tools sits at the low end. Tools that write to the system cost more, because each one needs limits, confirmation and testing for what happens when the AI calls it with the wrong input.
Access and discovery is its own line. Waiting on an IT team for API credentials, or working out an internal API with thin documentation, takes calendar time, not code. Quote it as a fixed fee so it does not quietly eat into the build.
Local or remote: the line that moves the price most
The MCP specification defines two standard ways a server talks to an AI app, and the one you build changes the scope more than anything else. Ask about it on the first call.
- Local (stdio). The AI app launches the server as a program on the user's own machine. The spec says these servers should take credentials from the environment rather than run a sign-in flow, so each user sets their own key. Simple to build and nothing to host, but every person has to install and configure it.
- Remote (Streamable HTTP). One server runs online and the whole team connects to it with a URL. In Claude Code that is one claude mcp add --transport http command per user. You now have hosting, uptime and security to handle, so price those in.
- Sign-in for remote servers. When a remote server needs authorization, the MCP spec bases it on OAuth 2.1. That is how each person signs in with their own account and only sees what their own permissions allow. It is real work, often the biggest single line, so never quote it as an afterthought.
A common middle path: a remote server with one shared service key, for a small team that is fine with everyone seeing the same data. Write that choice into the scope, so nobody assumes per-user permissions came included. More questions that prevent a bad quote are in client discovery questions before you quote.
Always quote the Keep plan
An MCP server is never finished on launch day. The client's API changes, people ask for new tools once they see the first ones work, AI apps update how they connect, and sign-in tokens and keys need rotating. A server that quietly breaks means the team goes back to doing it by hand and stops trusting the tool.
Quote a monthly plan in the original proposal. Say what it covers: monitoring that the server is up, fixes when the client's API changes, and a set number of small tool changes. Say what it does not cover: hosting and any paid API usage, which are billed at cost or held by the client, plus new tools and new systems, which are quoted separately. How to price a maintenance retainer covers the structure.
How to present the MCP server quote
- Lead with the job in their words. One line on what the team can now ask their AI to do, then the price.
- Itemize it. Access and discovery, tools, write safety, hosting and sign-in, then hosting costs and the Keep plan as separate monthly lines. Itemized quotes get trimmed, not rejected.
- Gate payments on proof. A deposit to start, a payment when the tools work end to end in Claude Code against a test account, and the final payment after the team is using it. How to structure milestone payments for an AI project covers the schedule.
- Write the scope down. Which tools, which can write, local or remote, shared key or per-user sign-in, and whose name the hosting and API accounts are in. How to write a statement of work for an AI project has the template.
And skip hourly. Claude Code makes the build fast, so hourly pricing turns that speed into a pay cut. Why that matters is covered in why I stopped charging hourly for AI builds.
Short, practical drops on skills, MCP, agents, prompts, and more. No spam, unsubscribe anytime.
Frequently asked questions
How much should I charge for an MCP server?
Price it on what it unlocks for the client's team: lookups, reports and updates they no longer do by hand. Charge a fraction of one year of that value for the build, itemize tools, write access, hosting and sign-in, bill hosting and API usage separately, and quote a monthly plan.
Is a remote MCP server more expensive than a local one?
Usually, yes. A local server runs on each user's machine and takes credentials from the environment. A remote server needs hosting, uptime and security, and if each user signs in with their own account, an OAuth 2.1 flow, which is often the biggest line in the quote.
Should MCP server tools be read-only?
Start read-only when you can. Tools that create, edit or delete records need limits, confirmations and extra testing, so they cost more to build and carry more risk. Add write tools once the client trusts the read-only version.
Who pays for hosting an MCP server?
The client. Have them hold the hosting account and any paid API accounts, or bill those costs at cost as their own line, separate from your build fee and Keep plan.
Should I charge a monthly fee for an MCP server?
Yes. The client's API changes, the team asks for new tools, AI apps update how they connect, and keys need rotating. Quote a monthly plan for monitoring, fixes and small tool changes, with hosting and usage billed separately.
Last reviewed by Duncan Rogoff on October 26, 2026
