The R2 Docuo AI connector — its MCP Server — lets an AI assistant you already use, such as Claude, ChatGPT, Gemini or Copilot, query the contents of your repository when you ask it to. You ask in plain language from your assistant and it gets the answer from your documents, without you having to open Docuo or export anything.
MCP stands for Model Context Protocol, an open standard that AI assistants use to connect to external services. Being a standard, the R2 Docuo connector is not tied to any one assistant: it works with any that speaks MCP.
The connector is switched off out of the box. Before you can connect your assistant, an administrator of your repository has to enable it. See Requirements and activation of the AI connector.
What you can ask it for
With the connector enabled, your assistant can:
- Find documents anywhere in the repository, by name or by what their file says.
- List a category, a folder or a saved query, with the columns you ask for.
- Work out totals, counts and averages over a whole set, without going through it document by document.
- Look up the workflow history: who moved a document, when, and what was sent back.
- Read what a document’s file says: Word, PDF, spreadsheets and scanned paper.
- Show you a document on screen: the file itself, not just its fields.
- Explain the design of the repository: its categories, fields, folders, saved queries and workflows.
On top of that come the tools of any packages you have installed and any your repository publishes on its own, if you have developed them. Your assistant will describe them if you ask what it can do.
In What you can ask the AI connector you will find many more examples and the full list of tools the connector publishes.
Example questions
- “How many invoices did we issue last month, and for what total?”
- “Give me the contracts expiring this quarter with their date and their customer.”
- “Who closed the August tickets?”
- “Find this customer’s file and summarise what the latest report says.”
- “What state is order 1043 in, and what can be done with it now?”
- “Show me the signed delivery note.”
How it works
- You authorise once. When you add the connector, your assistant takes you to an R2 Docuo screen where you sign in with your user name and password and choose the repository. Your password is typed into Docuo, never into the assistant.
- You ask in plain language. The assistant decides which tool of the connector it needs and what to call it with.
- The connector answers with your permissions. Every query is resolved as if you had run it yourself from Docuo: you see what you could already see, and nothing more.
- The assistant writes the answer from that data and shows it to you.
What your assistant can see, and what it cannot
The connector does not extend your permissions. It always operates within those assigned to you in the repository: it does not bypass the access restrictions configured there and it cannot reach repositories other than the one you authorised. If you cannot see a document in Docuo, neither can your assistant.
There are also operations no tool of the connector can perform, by design:
- Change your licence, the repository parameters, your balance or your billing details.
- Create, delete or modify users, collaborators or their permissions.
- Create or modify document categories, workflows or automations.
- Make payments, transfer funds or operate on payment methods.
- Sign on your behalf.
Collaborators have no access to the connector. It is a route reserved for the repository’s contracted users.
The tools your assistant offers depend on the configuration and the active packages of each repository, so those of one may not match those of another. Every tool is published marked as read-only, write or destructive; destructive ones require you to confirm expressly from your assistant before anything happens.
We will be adding new tools progressively. If one is withdrawn or changes in an incompatible way, we will give you at least thirty days’ notice, unless the change is necessary for security or regulatory compliance.
Security and privacy
Your authorisation is personal
The connector is authorised with your named account. The access token it generates is, to all effects, a service access password: do not share it with other people or across several accounts. Sharing it may lead to your access being revoked permanently.
Your assistant handles that token at the transport layer: the language model never sees it, it does not appear in the conversation and it cannot be leaked in an answer.
What gets logged
Every call is recorded in your repository’s audit log: the date and time, your user, the repository, the identification the assistant declares when it connects, the tool called, its parameters, whether it succeeded or failed, how long it took and the usage it generates.
- The content the connector returns is not logged.
- The log is stored in your own database, with the same security measures and the same separation as the rest of your repository’s information. It is part of the repository, and it is retained and deleted with it.
- Your conversation with the assistant never reaches the connector. All that reaches it are the tools it calls and the parameters it calls them with, which may reproduce part of what you asked.
What happens to the data once it leaves Docuo
When the connector answers, the content leaves R2 Docuo and is processed on the infrastructure of the assistant you chose. From that moment it is governed by the terms and the privacy policy you accepted with that provider, including those on retention, location and the use of data to train models.
The assistant and the model provider are independent third parties that you choose and contract freely. R2 Sistemas neither controls nor supervises them, and is not liable for their operation, availability, security or contractual terms. It is up to you to check that those terms are acceptable for the kind of information you keep in your repositories.
Two practical consequences worth being clear about:
- Check whether the plan you hold with your assistant uses your conversations to train its models and, if so, turn that off before you connect the repository. The business plans of the main assistants do not train on their customers’ data; some personal plans do.
- This is different from the AI features built into Docuo — the R2AI Assistant and the Artificial Intelligence Instructions — where the models are contracted by R2 Sistemas as data subprocessors and are contractually forbidden from training on your content. When you connect your own assistant over MCP, that guarantee comes from your agreement with it, not from ours.
Usage limits
The connector has its own usage limits, separate from those of the API. Each plan includes its own, and you can check them on the pricing page.
Once the included limit is used up, the excess can be covered with Credits for those limits the pricing page identifies as extendable that way. If your balance is insufficient at the time of the call, or the limit exceeded is not extendable, the tool does not run and returns an error; your assistant will tell you.
Bear in mind that a single question may trigger several calls. To answer “how much did we invoice per customer last month?”, your assistant may first look up the categories, then the fields, and finally the totals. Usage can be higher than working through Docuo by hand, so it is worth keeping an eye on your balance for the first few weeks.
Good practice
- Review answers before you use them. Language models are not deterministic: given the same question they may call different tools, with different parameters, and they may give answers that are wrong, incomplete or invented even when your repository’s data is correct. Verify the data and the citations you get against the repository, especially before using them in matters with legal, accounting, tax or employment effects.
- Answers are not advice and have no evidential value, and they do not replace the files or the evidence certificates in your repository, which always prevail.
- Be careful with files from outside sources. A file may contain text, visible or hidden, written so that a model reads it as an instruction. The connector delimits the content it returns so as not to give it instructional weight, but we cannot guarantee that an assistant will ignore it. Weigh this risk if your repository receives files from third parties.
- Start with one repository and a few people. Access is granted user by user, so you can try it with whoever is going to use it before opening it to the whole team.
How to connect
The process differs for each assistant, and each one changes it at its own pace. Pick yours:
If your assistant is not on the list but supports remote MCP connectors, you can connect it just the same: the R2 Docuo connector is a standard MCP server over HTTPS with OAuth authorisation, and any of the articles above gives you the server address and the details you need.
Most assistants register themselves and only need that address. Some — today Gemini and Copilot — also ask for a client id and client secret, which we issue through a support ticket at an administrator’s request; each connection article explains it, and so does Requirements and activation of the AI connector.
AI assistants move fast: their menus, their plans and the way connectors are added change frequently, and so does which of them needs client credentials. Each connection article states the date its procedure corresponds to.
References
- Terms and Conditions of Service: clause 16.2 (R2 Docuo MCP Server) and clause 21 (Artificial Intelligence Features).
- Privacy Policy.
- List of data subprocessors.
- Plans and pricing, for the usage limits included in each plan.