AI Agent Tool Permission Worksheet for ERP Access
Before an AI agent gets tools for your ERP, someone should be able to say exactly which actions it can take, on what data, and who signs off. This worksheet is that inventory, with a simple risk score per tool.
| Tool name | ERP or system | Access level | Data class | Impact (1-5) | Likelihood (1-5) | Risk score | Approval rule | Owner | |
|---|---|---|---|---|---|---|---|---|---|
4 | |||||||||
6 | |||||||||
10 | |||||||||
9 | |||||||||
12 | |||||||||
10 |
Suggested rule: any tool scoring 12 or more needs user confirmation, 15 or more needs manager approval, and anything touching export controlled or banking data should be reviewed by compliance before it is exposed. Review the sheet whenever a tool is added.
Review my agent permissions
Optional: send your completed sheet and we will flag gaps in roles, approvals and logging.
Optional. The tool and every export work without this. No spam.
How to use the worksheet
List one row per tool, not per agent, because the same agent will gain and lose tools over time. Set the access level, classify the data the tool can return, and score impact and likelihood from 1 to 5. The sheet multiplies them into a risk score. Then choose an approval rule: none for harmless lookups, user confirmation for drafts, manager approval for postings, and blocked where the data should never reach a model. Export to Excel for the audit file.
- Assign a named owner to every tool
- Separate draft writes from posting writes
- Re-score after incidents or schema changes
The method behind the score
Impact times likelihood is the standard qualitative risk matrix, kept deliberately simple so a plant IT team can run it in a workshop. The value is less in the number than in the conversation: why is the likelihood a 3, what would reduce it, and who is accountable. Least privilege means the ERP account used by the agent carries only the roles needed for the tools listed here, nothing broader.
Reading the example
The sample rows show a spread. Order status is low risk and only logged. Supplier bank details score high because of fraud exposure, so they are blocked. A purchase requisition draft needs the user to confirm, while a goods receipt posting needs a manager because it moves inventory and accounting. An export controlled drawing index is readable only after compliance review, which matters for ITAR or EAR obligations your company carries.
Common mistakes
Common faults are scoring by how impressive the demo is rather than by worst case, forgetting that a read tool can leak confidential data, sharing one powerful service account across all tools, and approving tools once without a review date. Keep the sheet current and tie it to change control so a new tool cannot ship without a row.
Frequently Asked Questions
Why score tools instead of agents?
Agents change prompts and capabilities often, but each tool is a concrete action with a defined data scope. Permissions and approvals attach naturally to tools, so the inventory stays accurate longer.
Is a read-only tool always low risk?
No. A read tool can expose pricing, bank details, personal data or export controlled technical data. Classify the data it returns and score impact on that basis, not on whether it can change records.
What approval rule should writes get?
Start with drafts that a person confirms, then posting actions that a role holder approves. Avoid autonomous posting until you have audit logs, limits and a track record from the draft stage.
How often should I review it?
Review when a tool is added or changed, after any incident, and at least quarterly. Tie it to your normal change control so it stays part of the process instead of a one-time document.
If you want help turning this inventory into enforced roles, approvals and audit logging in your ERP, Netray can build that layer on your infrastructure.
Related Tools
MCP Server Security Checklist
Work through authentication, authorization, input handling, prompt injection, secrets, logging and deployment before an MCP server touches ERP data.
Developer ToolsMCP Server Tool Definition JSON Generator
Fill in a tool name, description and one parameter to get a valid MCP tool definition with an inputSchema and behavior annotations.
Developer ToolsMCP Tool Catalog and Risk Worksheet
Register each MCP tool your agent exposes, the ERP object it touches and a computed risk score, then export to Excel.
Developer ToolsClaude Desktop MCP Config JSON Generator
Build the mcpServers entry for claude_desktop_config.json with the right command, arguments and environment variables.
More AI Agents & Automation tools
See all- Business Process Automation ROI Calculator
- AI Agent Use Case Prioritizer
- Document Processing Automation Savings Calculator
- AP Invoice Automation ROI Calculator
- Order Entry Automation Savings Calculator
- Quote-to-Cash Cycle Time Calculator
- Manufacturing AI Readiness Assessment
- RPA vs AI Agent Selector
- ERP AI Maturity Assessment