CSV Upload for MCP Servers
Some MCP servers are built for files you already download from partner portals, ad platforms, or finance tools that never shipped a live data connection. You upload on the MCPBundles dashboard; your AI reads what you stored through MCP tools in chat.
That flow is different from Webhook data capture (upstream systems POST to an ingest URL) and from Webhook triggers (wake an A2A agent).
| CSV upload (this page) | Webhook ingest | Webhook triggers | |
|---|---|---|---|
| How data arrives | You drop a file on the source in the side panel | An external system POSTs to an ingest URL | An HTTP POST starts an A2A agent run |
| Typical use | Weekly performance CSV, finance close, inventory export | Cron jobs, Zapier, form POST, monitoring hooks | Deal closed, payment received, form submitted |
| Listing example | CSV Analyzer | Webhook Ingest | Any A2A agent with a trigger URL |
When you see it
On a CSV Analyzer server, add a source. The side panel is where you name the source, write the note the agent reads, and drop .csv files or a .zip that wraps a single CSV.
The source panel asks for:
- Source name — how this export source appears in your workspace and in tool responses (one connection per property, brand, or portal you export from).
- Agent context (optional) — column meanings, date quirks, or KPI definitions your team wants the AI to remember.
There is no ingest URL for this pattern. Uploads always happen in the dashboard.
Setup flow
- Open CSV Analyzer (or another CSV-upload listing) in your workspace and add a source.
- Enter a Source name and optional Agent context, then Save.
- In the same panel, drop a portal export (CSV, or zip containing one CSV). Wait until the file appears in the list.
- Open Studio or install the server in Claude, Cursor, or Codex, then ask in natural language — for example: describe this file’s columns, or sum spend by campaign.
Each source can hold multiple uploads (for example one file per month when the portal only exports one period at a time), up to 24 files per source and 25 MB per file. New uploads add to the list; they do not remove older files unless you delete them. Use clear filenames (such as commissions-2026-01.csv) so you and the agent can tell months apart. A different property or portal is still another source. After upload, install the server in Claude, Cursor, ChatGPT, or another MCP client and ask in chat.
What your AI can do after upload
Capabilities depend on the listing, but CSV Analyzer typically lets the agent:
- List stored files on a source (filename, upload time)
- Check source status (name, file count, latest upload)
- Describe column names, types, and sample rows for a chosen file
- Run grouped calculations and filters (SQL-style analysis on the stored copy)
- Page through raw rows when exact cell text matters
When several files exist on one source, the agent should list them before assuming a single file, and target a specific upload when you ask about a particular month or filename. If you omit which file, tools use the most recently uploaded file on that source.
Uploads are read-only from the agent’s perspective — the model does not upload files for you. Add a new export by opening the source and dropping another file.
Tips
- One source per portal or property — add another source when the columns or the hotel change, so the note on each source stays accurate.
- Monthly downloads — keep separate files on the same source; name each file with the period (YYYY-MM) and mention that pattern in Agent context if helpful.
- Use Agent context for stable quirks (e.g. “Spend column is USD; dates are US format in Q1 exports”).
- Large files — analysis runs on a query-ready copy built at upload time; if describe or totals feel slow, ask for one metric at a time.