An MCP access without tools can do nothing. In the Tools tab you define what an assistant may do through this access – and this is the step that decides whether the access later works briskly and accurately or sluggishly and imprecisely.
The “Tools” tab
Configuration → Interfaces → MCP configurations → <open the access> → Tools
Each row of the list shows: No., Label (the name the assistant sees), Base tool (the underlying tool from the construction kit), Language and Active.
On the left, the details panel shows the decisive figure under Details:
Tokens (estimated): 42.2k tokens (fits comfortably)
Why the token estimate matters
It is not a statistic but a warning light.
An assistant is handed the complete tool list when it connects, and it has to keep it in mind on every single question. That list occupies space in the model’s working memory – measured in tokens. The more tools, the larger the list, the less room remains for the actual task, and the longer every reply takes.
On top of this comes a second, more unpleasant effect: the more similar tools there are to choose from, the more often the assistant reaches for the wrong one. Twenty tools, six of which do “something with tickets”, are an invitation to mix them up.
The token estimate makes both visible before a user complains. The access above sits at 42.2k tokens and “fits comfortably” – so the estimate gives the all-clear. Even so, at this size it is worth asking whether all the tools are really needed, or whether two accesses would make more sense here: “fits” means it works – not that it is optimal.
Rule of thumb: Two accesses with twelve fitting tools each beat one with twenty-four – even if the same people use both. It costs you one more click in administration and saves every user time on every question.
Adding a tool
- Open the MCP configuration and switch to the Tools tab.
- Click the plus at the top right. The MCP Tool dialog appears.
- Choose the Category (mandatory field). It groups the tools by area – each category carries a short explanation, e.g. “Tools for managing open items”.
- Choose the tool you want and save.
Which categories and tools exist is covered in Available MCP tools.
How to tailor it
Tailoring works best when you think from the task, not from the catalogue. Not “what is there?”, but “what do these people do all day?”.
- Write down the use case. One sentence: “The sales team wants to analyse contacts and sales opportunities.”
- Take the tools that sentence needs. For “analyse contacts”, Search contacts and Contact information belong in – expenses don’t.
- Add “Who am I”. It costs next to nothing and helps the assistant understand questions like “my tickets” at all.
- Look at the token estimate. If it stays comfortable, you’re done.
- Check write tools one by one. Each of them is a deliberate decision, not a by-catch.
A real-world example: an access “Employee” with the description “For everyday tasks” gets by with 36 active tools and around 42k tokens – Agenda, Who am I, Activities, Employees, Contacts, Tickets, Projects, Organisations. That covers the everyday work of most colleagues, without a personnel or finance tool ever being involved.
Common questions & needs
| You want to … | How to |
|---|---|
| Make an access faster | Remove tools that don’t come up in the use case. Check via the token estimate in the details panel. |
| Build a read-only access | When adding, choose only querying tools (search, retrieve information) and no write tools. |
| Temporarily disable a tool | Remove the Active flag on the tool’s row instead of deleting the tool. |
| Know how large the access is | Details panel on the left → Details → “Tokens (estimated)”. |
| Have the assistant understand “my tickets” | Include the Who am I tool – without it the assistant doesn’t know the logged-in user. |
| Make an assistant reach for the wrong tool less often | Fewer similar tools in the same access. When in doubt, split into two accesses. |
| Give a tool a clearer name | Adjust the Label – it is what the assistant reads. The Base tool stays unchanged. |
Notes
- Leaving tools out is not a security measure. What a user may see is decided by their permission profile – see Security and permissions with MCP.
- The label is the text the assistant reads. Meaningful names measurably improve the hit rate; cryptic abbreviations worsen it.
- Changes take effect immediately. On next connecting, the client sees the new list; no one has to set up their connector again.
- The language sits with the configuration. All tools of an access speak the same one – for mixed-language teams, create two accesses.
Related topics
- Create an MCP configuration MCP & AI agents Configuration
- Available MCP tools MCP & AI agents Reference
- How MCP works in teamspace MCP & AI agents Concept
- Security and permissions with MCP MCP & AI agents Concept