One API, 3 tools
$500
5 days · 1 revision
A first MCP server for one API or one feature of your app.
- MCP server for one API or feature
- Up to 3 tools
- Token authentication
- Source code in your repository
Fixed-price project
I build an MCP server for your Laravel app or API, so AI agents can use named tools with token login, limited permissions and a log of every call. It starts at $500 for one API and 3 tools, in 5 days.
Who it is for
Signs
Scope
We pick the few things the agent should do first, such as look up an order or search customers. Each one becomes a small named tool with clear inputs. I keep tools narrow, because an agent can only misuse what you gave it.
Every call needs a token, and each token gets only the permissions its job needs. I check permission inside each tool, not only at the front door. The Laravel MCP package supports Sanctum tokens and OAuth through Passport. Token login is what the packages include. If your client needs OAuth, ask me before you order.
Reading is easy to make safe. Writing needs more care. A tool that changes data gets its own permission, input checks and a log entry. Tool annotations such as read-only or destructive are untrusted metadata, so I enforce the rules in code. Where a mistake would be costly, I suggest a draft or a person to approve.
The middle package adds scoped access and logs of calls. The largest adds rate limits and monitoring. The MCP spec says servers must validate tool inputs, check access and rate limit calls.
Access to the Laravel app or API (a repository invite or a zip file), the three things you want an agent to do first, and the AI client you plan to use, such as Claude. A staging copy of the app helps. I do not need live production keys.
The AI model and its usage bills are yours. The smaller packages deliver a working MCP server, not a finished chatbot or agent. Only the largest package includes an AI agent on top. I cannot promise an agent will never be fooled by text it reads, so the design limits what it can do.
Packages
Fixed price in US dollars. Each package lists what is included.
$500
5 days · 1 revision
A first MCP server for one API or one feature of your app.
$1,200
12 days · 2 revisions
You need a wider set of tools, with different access for different jobs.
$2,500
21 days · 3 revisions
You want a larger tool set, with limits and monitoring, and an agent working on top.
Process
Step 1
You give me access to the Laravel app or API and tell me the three things you want an agent to do first.
Step 2
I propose the tool list, what each tool takes and returns, and what each token may do. You approve it before I build.
Step 3
I add the MCP server to your Laravel code, with token login and the permissions we agreed, and connect it to a real MCP client.
Step 4
You receive the source code, a connection guide for Claude, and a walkthrough call.
Proof
Article
Practical AI agent security for Laravel apps and MCP servers. Learn how to limit prompt injection damage with scoped tokens, approvals, and...
Article
I built a Laravel MCP server that adds HubSpot Lists support missing from the native connector, and reused two of my own Guzzle observabilit...
Article
Your AI agent can read your whole codebase but not your production errors. I built a free Laravel package that fixes that in a few minutes.
Article
A REST API serves code that developers write. An MCP server serves AI agents that pick tools while they run. I explain the difference, when...
FAQ
It is a small service that tells an AI app what it can do in your system. MCP is an open standard, and apps such as Claude, ChatGPT and Visual Studio Code can connect to it. Your app offers named tools, and the agent picks among them while it works.
It can be, if the agent can only reach what you chose. I cannot promise an agent will never be tricked by text it reads, so I limit the damage instead. That means narrow tools, one token for each job, permission checks in code, and logs. Risky actions can wait for a person.
Yes, if you allow it, and only through tools you approve. Reading is the safe default. For writes I add input checks, a separate permission and a log entry. For costly actions I suggest a human approval step, or saving changes as drafts first.
Yes. You get the full source code for the server in your repository, so your team can add or change tools later. The setup guide and the short walkthrough show how the server connects to a client such as Claude and how a new tool is added.
Any app that supports MCP. I write the setup guide for clients like Claude, and the MCP site lists ChatGPT and Visual Studio Code among the apps that can connect. Tell me which one you plan to use, and I will tell you if I see any limit before you order.
An API is made for code that a developer writes, with fixed calls. An MCP server is made for an AI agent that picks tools while it runs, and it tells the agent which tools exist. Often the MCP server sits on top of your API or your app code, so your logic is not copied.
Yes. I run an MCP server on my own production site. I also tested an open-source Laravel Pulse MCP package in my own production app before sharing it. I have built one for HubSpot Lists too, and I wrote about both and about MCP security for Laravel apps.
If your work does not fit these packages, tell me what you need and I will reply with a plan.
More projects
From $250
Set up Claude Code as a small team, with your rules, sub-agents and skills.
From $250
A senior engineer checks your AI-built app before you launch it.
From $300
Contacts, companies and deals kept in step between Laravel and HubSpot.