ServiceSoftware development
Claude and MCP integrations: give your team's AI safe access to your own systems
Our Claude and MCP integration service builds a connector — an MCP server — that lets the AI assistants your team already uses, such as Claude on claude.ai or Claude Code, read and act on your company's own systems: CRM, projects, documents, orders. Each person connects their own assistant with their own access, sensitive actions need approval, and every call is logged. We run exactly this for our own team.
What we do for you
- A map of what the AI should be able to read and do in your systems — and what it must never do
- An MCP server exposing a small set of well-described tools (search a client, create a task, read a document…)
- Per-person keys or sign-in, so each assistant acts with that person's rights
- Approval steps for anything that changes data or sends something outside
- A log of every call: who, which tool, what input, what result
- Hosting with the usual foundations (dev, uat, prod, builds on GitHub Actions, Docker, health checks)
- A short guide for each team member on connecting their own Claude
Who it is for
- Fits: teams already using Claude (or another MCP-compatible assistant) who copy-paste data from internal tools into it
- Fits: companies with an internal tool, CRM or database the AI could query instead of a person
- Fits: managers who want AI access to company data to be controlled and traceable rather than ad hoc
- Does not fit: giving an AI unrestricted write access to production systems — we will not build that
What MCP is, in one paragraph
The Model Context Protocol (MCP) is described by its maintainers as "an open-source standard for connecting AI applications to external systems" — they compare it to a USB-C port for AI applications. An MCP server offers tools, resources and prompts; an AI application such as Claude or ChatGPT connects to it and can then use those tools when the user asks. Build the connector once, and any compatible assistant can use it.
What we built for our own team
Our team works in an internal operating system: CRM pipeline, work items, roadmap, priorities, documents. We built an MCP server on top of it so that each person can plug their own Claude into it — on claude.ai or in Claude Code — and ask in plain language: "what is open on this client?", "create the follow-up task", "show me this week's priorities". The server checks who is calling, applies that person's rights, and logs every call. The same assistant can therefore help with real company data without anyone pasting it into a chat.
Security: the rules from the protocol and from Anthropic
The MCP specification is explicit: "Tools represent arbitrary code execution and must be treated with appropriate caution", hosts "must obtain explicit user consent before invoking any tool", and users must keep control over what data is shared and what actions are taken. It also notes that the protocol cannot enforce these principles itself — implementers must build consent, authorisation and access controls into their applications.
Anthropic warns that custom connectors let Claude access "and potentially modify data within those services based on your permissions", and advises to "only connect to trusted servers" and to beware of prompt injection: malicious servers "may include hidden instructions". Claude Code's documentation gives the same warning. On Claude Team plans, Anthropic says only Owners and Primary Owners can add custom connectors.
- Few tools, each doing one clear thing, with honest descriptions
- Read-only by default; write tools only where needed, with approval for anything sensitive
- Per-person authentication, so the AI never has more rights than the person using it
- No tool that fetches arbitrary external content into your systems
- Every call logged and reviewable
How we build it
- Pick the questions. The ten things your team most often looks up or does in your systems.
- Design the tools. One tool per action, with clear inputs and limits. Write tools reviewed with you.
- Build and test. Against a copy of your data in dev and uat, with real questions from the team.
- Connect. Each person adds the connector to their own assistant; on Team or Enterprise plans an owner adds it for the organisation.
- Watch the logs. We review which tools are used, which fail and which are missing, and adjust.
MCP connector, chatbot or API integration?
| Option | Best for | Limit |
|---|---|---|
| MCP connector | People using their own AI assistant on company data | Needs assistants that support MCP and careful permissions |
| Custom chatbot | Customers or staff without an AI assistant | A separate interface to build and maintain — see website AI chat |
| API integration | Data that must flow automatically between systems | No conversation — see API integrations |
Questions we get
Do we need a specific Claude plan?
Anthropic says custom connectors using remote MCP are available on its Free, Pro, Max, Team and Enterprise plans; on Team plans only owners can add them. We check your setup before building.
Can it work with assistants other than Claude?
MCP is an open standard supported by several AI applications, including ChatGPT and development tools. We test with the assistants your team uses.
Can the AI delete or change our data?
Only through tools we deliberately expose, with the rights of the person using it, and with approval for sensitive actions. By default, tools are read-only.
Is every action traceable?
Yes. Every call is logged with who made it, which tool, the input and the result.
Who owns the connector?
You do. The code is in your repository and the server runs on hosting you control or that we manage in your name.
Sources
- Model Context Protocol — What is MCP? (checked 2026-10-06)
- Model Context Protocol — Specification (Security and Trust & Safety) (checked 2026-10-06)
- Claude Help Center — Getting started with custom connectors using remote MCP (checked 2026-10-06)
- Claude Code Docs — Connect Claude Code to tools via MCP (checked 2026-10-06)