User and agent permissions
We define resources and operations for the user and the agent’s role. We specify whose authority the agent uses, and the destination system also verifies access.
Data and context · Integrations
AI that works in company systems.
From checking availability in ERP to preparing a quote in CRM. We give agents the data and operations needed for a specific task, with an agreed scope of access and control.
MCP
MCP connects AI to tools. The company sets the scope.
Model Context Protocol (MCP) is an open standard for exposing tools and context to AI applications. It describes the resources and operations an assistant or agent can use.
AI EngineModel Context Protocol
- 01
People and agents
The task, identity and role in the process determine which access is needed.
- MCP
Tools and context
An MCP server exposes selected resources and operations. Each has a defined purpose, inputs and results.
- 03
Company systems
A connector or API executes the permitted operation and returns the result to the agent.
Access scope · approvals · execution records
Data stays in its source systems or in Context Lakehouse. MCP exposes the required context and operations, while each connection must fit the interface and rules of the particular system.
Systems
One task. Several systems.
An agent needs access to a specific piece of information or an action. We design the connection around that purpose, starting with the systems the team already uses.
ERP
Products, availability, prices and order status. We agree the scope of reading and any writing with the process owner and system administrator.
CRM
Customer context, a sales opportunity and contact history. An agent can prepare a record update or a draft quote, following the agreed approval path.
Microsoft 365
Email, files and work in Teams. We expose selected resources and operations according to identity, permissions and organisational policies.
Helpdesk
A request, attachments and the history of a ticket. The connection can support drafting a response, completing information or handing the case to the right person.
APIs and legacy systems
We use available interfaces or an agreed intermediary. Where no API is available, we assess a possible access path, such as a read-only replica, and match the agent’s scope to it.
Example
From a request to an approved quote.
An example Quoting agent workflow connects the helpdesk, ERP and CRM. The agent gathers information and prepares a draft. Process rules determine when it may send or save it.
Explore the Quoting agentExample workflow · Quoting agent
- 01
Read
The agent retrieves the request and attachments from the helpdesk. It checks permitted product data, prices and availability in ERP.
- 02
Draft
A draft quote is linked to the customer and its data sources. Missing information returns as questions for the person responsible for the case.
- 03
Approval
The responsible person reviews the scope, terms and recipient. Operations requiring approval wait for that decision.
- 04
Write and confirm
After approval, the connector performs the permitted operation in CRM or the helpdesk. The process checks the result and records the change identifier.
Control
Know who is acting. Know what happened.
Control is part of the system connection. We agree the rules with the process owner and IT, then verify them against real work scenarios.
Reading and approved operations
We distinguish retrieving information from changing data, sending messages or starting a process. The company defines autonomous actions and approval points.
Logs and observability
We record calls, identity, timing, results and approval decisions. We monitor errors and latency. Log scope, access and retention are agreed with IT.
For errors and uncertain write results, we define a path to check status, retry or hand over to a person. This protects the process from uncontrolled repetition of an operation.
First step
Start with a task that comes back every day.
We choose one path from information to an outcome. Further tools and systems are added after the first workflow has been verified.
- 01
Process and outcome
The process owner identifies the task, exceptions and criteria for successful completion. We define what the agent should read, prepare and write.
- 02
Access and rules
IT confirms interfaces, accounts, permissions and a test environment. Together, we specify operations that require approval and how they will be supervised.
- 03
Verify in practice
We test the whole workflow, access denial, errors and retries. Then we expand the scope to further tasks.
Time saved
Less manual data transfer and switching between systems to complete the same task.
Sales and revenue
Faster quote preparation with available customer, pricing and availability data.
New possibilities
One agent can carry a task across several systems, within agreed permission boundaries.
Let’s start with one process and the systems it needs.
Book a callLet’s talk about the work you want to improve.
Tell us how a process works today. Together, we will explore where AI Engine can help and which module to start with.
Book a callOr write to: hello@wearefuture.ai
