Scenario 90: Running the monthly report folder with Claude Code and extending MCP

Who and in what situation

İrem, the factory’s operations analyst, combines production CSVs each month and creates a management summary in a secure folder prepared by IT. She does not write code; IT sets up the folder structure, permissions and MCP connection.

How it is done today

İrem checks file names, merges columns manually, copies the report and prepares an email draft. When the format changes, she asks IT for support.

How to do it with AI

1. IT sets up the internal data source as an MCP server with the necessary tools. For a remote server, claude mcp add --transport http {NAME} {URL} can be used; the connection is checked with /mcp. [150]
2. İrem opens the Code tab in Claude Desktop or the terminal in the secure folder. Claude Code can read files and run commands; the approval rule remains in place. [160]
3. Put the report prompt in a script in the format claude -p "{TASK}" --allowedTools "Read,Edit,Bash" --output-format json. -p runs non-interactively. [162]
4. IT can add a hook that formats output with PostToolUse or provides a task completion alert with Notification. [163]
5. Check the source files, date range and totals before publishing.

Example prompt

Read the {MONTH} CSV files in {REPORT_FOLDER}. Compare the column names first; list missing or different columns separately. Generate {REPORT_NAME}.xlsx and a brief executive summary only in {OUTPUT_FOLDER}. Do not modify the source files, send data to external systems or publish uncertain totals without approval.

Evidence and real-world example

Salesforce’s use of Claude Code and Slack MCP is a published example. The same MCP server approach is also supported through the Plugins > Add custom MCP server flow in ChatGPT Business and Enterprise/Edu and the Tools > Add a tool > New tool > Model Context Protocol flow in Copilot Studio. [158] [164] [165] [166]

Points to consider

Claude Code is a terminal and file tool; for non-developer users, IT needs to set up the folder structure and permissions. Do not connect untrusted servers, review tools with write access and start in read-only mode. OpenAI and Microsoft also emphasize trust and administrator control. [146] [160] [164] [165]

Sources for this scenario

146. Get started with custom connectors using remote MCP (support.claude.com)
150. Connect Claude Code to tools via MCP (code.claude.com)
158. Salesforce and Anthropic expand partnership (anthropic.com)
160. Claude Code Security (docs.anthropic.com)
162. Run Claude Code programmatically (code.claude.com)
163. Automate actions with hooks (code.claude.com)
164. Connecting custom MCP servers in ChatGPT (help.openai.com)
165. Connect your agent to an existing Model Context Protocol server (learn.microsoft.com)
166. Extend your agent with Model Context Protocol (learn.microsoft.com)