MCP server development services: what to buy and how to check it
A custom MCP server sits between your systems and an AI client and decides what the model may read and do. Vendors quote two to ten weeks for a production build and price it in very different ways. This guide covers what you are buying, what the ranking vendors publish, the security checks and the questions to ask.
Buyer guide, MCP
- Checked
- 10 November 2026
- Typical launch
- 4-8 weeks at WiserBrand
- Per engineer, monthly
- $7,000 or more at Ryz Labs
- Project minimums
- $10,000 in Entrans's list
- Spec version read
- MCP 2026-07-28
The short answer
A custom MCP server exposes tools, resources and prompts from your systems to AI clients over stdio or Streamable HTTP, with authorization when it serves users. Vendors quote two to ten weeks. Published prices include $10,000 project minimums in Entrans's list and $7,000 or more per engineer each month at Ryz Labs. Test auth and tool output before you sign.
What a custom MCP server is
The MCP project's Understanding MCP servers page, for the 2026-07-28 version, gives a server three building blocks. The architecture page adds two transports, and the authorization guide adds access control.
- Tools. Functions the model decides to call, each with a JSON Schema for its inputs. They can write to databases and call APIs.
- Resources. Read-only data such as files, database schemas and documents, chosen by the application.
- Prompts. Reusable instruction templates that the user picks.
- Transports. A stdio server runs locally and usually serves one client. A Streamable HTTP server is remote and usually serves many.
- Authorization. Optional, and strongly recommended when the server handles user data or must audit who did what. It follows OAuth 2.1, with your server acting as a resource server that validates tokens.
Build, adopt or skip
Start by asking whether you need a server at all. Provanta says a Zapier flow or a webhook is faster and cheaper when the job is a fixed trigger and action, and that MCP pays off when the AI chooses which tool to use.
Adopt an existing server when one covers your system. Anthropic's Claude Code docs say to verify you trust each server first, because servers that fetch external content can expose you to prompt injection.
Build when the capability is yours: your rules, your data, your identity provider. Ryz Labs wraps an internal API as a small set of well-described tools, not a one-to-one copy of every endpoint. Order Group builds a thin layer so validation stays in the API.
What the ranking vendors publish
Every row comes from the vendor's own page, read on 10 November 2026.
| Vendor | Scope published | Timeline | Price | Proof shown |
|---|---|---|---|---|
| WiserBrand | Strategy, build, integration, auth, security, deployment and support | Typical launch 4-8 weeks, in five stages | None published | A Magento server, 80% less search time |
| Provanta | Servers and clients in TypeScript, Python and .NET, with scoped permissions and audit logging | 2-4 weeks for one or two systems, 6-10 for complex ones | None published | Says upfront when MCP is the wrong choice |
| Ryz Labs | Pods of senior engineers: internal API servers, OAuth remote servers, gateways | Threat model and first local server in week 1 | $7,000 to $15,000 per engineer per month, about $40,000 a month for a starter pod, about $120,000 over three months | Names the people and the price before you start |
| Order Group | Tools over your existing API, token auth, call log, server-side gates | Not published | None published | Runs its own server since July 2026. Its FAQ says "Not yet" to client MCP projects |
| Entrans list | Ten vendors, Entrans first | Not published | $20 to $120 an hour, $10,000 to $25,000 minimums, platform costs on top | Ranks itself |
| Phosai Labs list | A ranked list of vendors | 2-4 weeks simple, 4-8 production, 3 months or more at enterprise scale | None published | Names its own implementation partner first |
How to read the rankings and quotes
Lists. Entrans and Phosai Labs each publish a ranking. Entrans places itself first, and Phosai names the firm it works with. Treat both as vendor pages and check each name yourself.
Timelines. Vendors agree on the shape: two to four weeks for a simple server, four to eight for production, longer with compliance. Ryz Labs says timelines depend on scope and your security review, so ask which a quote includes.
Prices. WiserBrand, Provanta, Order Group and Phosai Labs publish no price. Ryz Labs prices per engineer per month and the Entrans list per hour. Ask for a fixed price against a named scope.
Security checks a production server needs
The MCP project's own guides set the baseline. Ask a vendor to show each check working.
- Authorization. A remote server validates tokens as an OAuth resource server and answers an unauthenticated call with 401 and a WWW-Authenticate header.
- No token passthrough. The security guide calls forwarding a client token to a downstream API an anti-pattern. The server accepts only tokens issued to it.
- Narrow scopes. The guide lists wildcard scopes as a common mistake and warns against trusting scope claims without server-side authorization.
- Tool output as untrusted text. WiserBrand lists prompt and tool-injection testing. Order Group returns HTTP 400 or 409 when an action breaks a rule, so the rule lives on the server.
- Logging. Record who called which tool and the result, with no sensitive data. A stdio server must log to stderr, because output on stdout corrupts the protocol.
- Proxy and fetch controls. A server that fronts a third-party API needs per-client consent, and one that fetches URLs needs SSRF protection.
Scoping checklist and questions for a vendor
Settle these six points before you ask for a quote:
- Which AI clients will connect: Claude, ChatGPT, Cursor or your own.
- Which systems, and which tools read or write.
- Whose identity each call runs as.
- Local or remote hosting, and who operates it.
- Who approves a new tool or scope later.
- What done means: tests, tool-selection evals and a threat model.
Then ask the vendor these:
- Can I see a server running in production, and who calls it?
- Do the rules live on the server or in the prompt?
- Which transport and auth flow, and how are tokens scoped?
- Do tool calls land in a log that I own?
- Who owns the code and the cloud accounts at handoff?
- How do you handle protocol and client changes after launch?
Where Digiton fits
We build MCP servers as part of production work for teams that already use Claude:
- A server that gives a new agent its tools goes through AI agent development.
- A server inside a larger system goes through custom AI development.
- A server a vendor already built goes through our production engineering review, which will give you a scorecard across ten layers with evidence for each.
Our MCP security guide goes deeper on the attacks, and our Claude agent harness guide covers the loop that calls your server.
Frequently asked questions
What does an MCP server do?
It exposes tools, resources and prompts from your systems to AI clients through the Model Context Protocol. Your existing APIs and databases stay in place, and the server decides what the model may reach.
How long does a custom MCP server take to build?
Vendors quote two to four weeks for one or two systems and four to ten weeks for production builds with custom auth. Provanta, WiserBrand and Phosai Labs publish these ranges, and Ryz Labs says it depends on scope and your security review.
Do I need authorization on an MCP server?
The protocol makes it optional. The MCP authorization guide strongly recommends it when the server handles user data, needs an audit of who did what or serves enterprise users.
Sources, checked 2026-11-10
11 sources
- The MCP project's documentation gives servers three building blocks: tools the model calls, read-only resources and prompt templates.Source: modelcontextprotocol.io, checked 2026-11-10.
- Servers using the stdio transport typically serve a single client, while remote Streamable HTTP servers typically serve many, and new implementations should log to stderr or use OpenTelemetry.Source: modelcontextprotocol.io, checked 2026-11-10.
- MCP authorization is optional, follows OAuth 2.1 and treats the MCP server as a resource server that validates tokens.Source: modelcontextprotocol.io, checked 2026-11-10.
- The MCP security guide calls token passthrough an anti-pattern and lists wildcard scopes and unverified scope claims as common mistakes.Source: modelcontextprotocol.io, checked 2026-11-10.
- Anthropic warns that MCP servers that fetch external content can expose you to prompt injection.Source: code.claude.com, checked 2026-11-10.
- WiserBrand states a typical MCP server launch of 4-8 weeks in five stages and lists prompt and tool-injection testing.Source: wiserbrand.com, checked 2026-11-10.
- Provanta quotes 2-4 weeks for a server connecting one or two systems and 6-10 weeks for complex setups, and says a Zapier flow or webhook can be the better choice.Source: provanta.ai, checked 2026-11-10.
- Ryz Labs publishes $7,000 to $15,000 per engineer per month and a starter pod of about $40,000 a month.Source: ryzlabs.com, checked 2026-11-10.
- Order Group says it has run its own production MCP server since July 2026 and has not yet built one for a client.Source: ordergroup.co, checked 2026-11-10.
- Entrans's list of ten vendors places Entrans first and states rates of $20 to $120 an hour with $10,000 to $25,000 project minimums.Source: entrans.ai, checked 2026-11-10.
- Phosai Labs quotes two to four weeks for a simple MCP server and four to eight for a production-grade one.Source: phosailabs.com, checked 2026-11-10.
Digiton Dynamics OÜ is registered in Tallinn, Estonia, and delivers from Lisbon.
Update log
- Published. Read the WiserBrand, Provanta, Ryz Labs, Order Group, Entrans and Phosai Labs pages and the MCP project's server, architecture, authorization and security pages.
Each source above was opened on the date beside it. How facts get chosen and checked is written down in the editorial policy.
Talk to an engineer about your system
Book a free 25-minute scoping call. We tell you what we would build, what it takes to keep it running, and whether we are the right fit.