MCP server development
MCP servers that let AI assistants work inside your platform
The Model Context Protocol (MCP) is how Claude, ChatGPT, Cursor and other AI clients connect to outside systems. We build MCP servers for SaaS products and internal platforms — remote, authenticated, permission-aware and maintained like the rest of your API.
Why companies are building MCP servers now
Your customers and staff increasingly work inside AI assistants. If your product is not reachable from those assistants, someone will copy data in and out by hand — or pick a competitor that is. An MCP server makes your platform a first-class tool for any MCP-compatible client.
Most public MCP examples are local scripts with a single API key. A production server for a real business needs remote hosting, proper sign-in, per-user permissions, rate limits and an audit trail. That is what we build.
An MCP server we built and run in production
For a US leadership-development consultancy, we built a remote MCP server inside the admin platform we had already developed for them. Staff connect Claude in one click and run operations — invoices, calendar, follow-ups, time entries, contracts, HR and compliance tasks — through roughly 115 tools, each limited to what that staff member may do in the app.
What a production MCP server needs
Remote, not local
A hosted endpoint (Streamable HTTP) your users add by URL — no local install, no API keys pasted into config files.
Standards-based sign-in
OAuth 2.1 authorization as the MCP spec describes, so each connection acts as a specific user and access can be revoked.
Tools mapped to your permissions
Tool availability and behaviour follow the user's role. Read-only users get read-only tools.
Thin tools over your real API
Each tool calls your existing validated endpoints, so business rules are enforced in one place and the server stays small.
Tool design for models
Clear names, tight input schemas and compact, useful outputs — the difference between a model that picks the right tool and one that flails.
Audit and limits
Every tool call is logged against the user, with rate limits and confirmations on destructive actions.
What we build
Public MCP servers for SaaS products
Let your customers use your product from Claude, ChatGPT and other clients — a distribution channel as much as a feature.
Internal MCP servers
Give staff AI access to your admin systems, CRM, ERP or databases with the same permissions they already have.
API-first groundwork
If your platform lacks a clean API, we build that first so the MCP server (and everything else) has something solid to call.
Upgrades of local prototypes
Turning a working stdio prototype into a hosted, authenticated, multi-user server.
How we deliver an MCP server
- 1
Map the jobs, not the endpoints
We list what users actually want to do from an assistant and design tools around those jobs, rather than exposing every endpoint one-to-one.
- 2
Design auth and permissions
Decide how users sign in, which roles see which tools, and which actions need confirmation.
- 3
Build and test with real clients
We test tool selection and outputs in Claude and other MCP clients against realistic tasks, and tighten tool descriptions until the model behaves.
- 4
Deploy and document
Hosted alongside your API, with monitoring, usage logs and setup instructions for your users.
Integration and deployment considerations
Where it lives
Inside your existing backend (as we did in a Next.js app) or as a small separate service in front of your API.
Spec changes
MCP is evolving quickly. We track the specification and SDK releases and keep your server compatible.
Client compatibility
We test against the clients your users actually run — Claude, ChatGPT, Cursor, VS Code and others.
Security review
Prompt-injection exposure, over-broad tools and data leakage are reviewed before launch, not after.
Technology we use
Protocol & SDKs
- Model Context Protocol
- MCP TypeScript SDK
- Streamable HTTP transport
Auth
- OAuth 2.1
- Dynamic client registration
- Your existing identity provider
Runtime
- Next.js / Node.js
- Vercel
- Google Cloud Run
Who this is for
- SaaS companies whose customers want to use the product from AI assistants.
- Businesses that want staff to run internal systems through Claude or ChatGPT safely.
- Teams with a local MCP prototype that need it hosted, authenticated and multi-user.
Further reading from our engineers
- Building a production MCP server for your SaaS: auth, tool design and lessons from ~115 tools
What it actually takes to put a Model Context Protocol server in front of a real product: remote transport, OAuth, permission scoping, tool design and testing with real clients.
Need an engineer inside your own team instead of a project? Hire AI developers from DablewDevs on an hourly or dedicated basis.
Related services
MCP Server Development: common questions
Tell us about the platform you want to connect
Tell us what you are building or what needs fixing. You will hear back from an engineer — not a sales rep — within one business day, usually with questions or a rough approach.
Your enquiry goes straight to our engineering team — the people who will scope and build your project, not a sales desk.