MCP and automation
CSV Dashboards in MCP and Agent Workflows
MCP-style workflows can make a CSV dashboard available where people already ask questions. The important part is a clear contract: what the tool can read, what it can change, which data is returned, and how access is revoked.
Upload
Bring the CSV you already use.
Understand
Start with the brief and evidence.
Act
Ask, edit, filter, and export.
Define small, explainable tools
Start with focused operations such as list datasets, profile a CSV, ask a question, export a view, or update an authorized workspace. Small tools are easier to test and safer to expose than one unrestricted command.
Separate read from write
Reading a profile and editing a cell should not share the same permission. Use explicit scopes, confirmation for writes, audit logs, rate limits, and a revoke path for every token.
Make the agent honest
Tool descriptions should state the dataset and columns available, the freshness of the result, and the limits of the analysis. If an agent cannot prove a claim from the CSV, it should say so.
Common questions
Can I connect I.S.A.A.C to an automation platform?
Yes, the documented HTTPS API and scoped tokens are designed to support automation. Follow the current API documentation for the available endpoints and limits.
Should an agent be allowed to edit CSVs automatically?
Only with a separate write scope and a user-confirmed workflow. Read-only analysis should be the default.
Keep exploring
See what your own CSV is saying.
Try the browser workspace with a real file and keep the public guide open when you need help.
Start free