Load, query, filter, and analyze CSV files with pandas via csv-mcp-server (uvx)

csv-mcp-server — pandas-powered CSV operations via MCP

Package: csv-mcp-server (PyPI) — aka "CSV Editor" Launch: uvx --from csv-mcp-server csv Transport: stdio, NDJSON framing (newline-delimited JSON, NOT Content-Length) Auth: none required Server info: CSV Editor v3.4.2 (FastMCP 3.4.2)

Tools (38 total)

Loading: load_csv (file path), load_csv_from_url, load_csv_from_content (inline string) Inspection: get_session_info, list_sessions, get_statistics, get_column_statistics, get_correlation_matrix, get_value_counts, profile_data Transform: filter_rows, sort_data, select_columns, rename_columns, add_column, remove_columns, change_column_type, update_column Clean: fill_missing_values, remove_duplicates Analysis: group_by_aggregate, detect_outliers, find_anomalies, validate_schema, check_data_quality Export: export_csv (csv, json, etc.) History: undo, redo, get_history, restore_to_operation, clear_history, export_history Auto-save: configure_auto_save, disable_auto_save, get_auto_save_status, trigger_manual_save Meta: health_check, get_server_info, close_session

Session model

The server is session-based: load_csv returns a session_id that you pass to all subsequent operations. Data mutations (filter, sort, add_column) modify the session in-place. Use undo/redo to navigate history.

Gotchas

1. NDJSON only — Content-Length framing is ignored. Send one JSON object per line.
2. filter_rows mutates — filtering reduces the dataset in the session. Subsequent operations work on the filtered subset. Use undo to restore.
3. group_by_aggregate may fail after filter — if you filter to a single group then try to groupby, the server returns `{"success": false, "error": "GROUPBY"}`. Reload or undo first.
4. Executable name is `csv` — NOT csv-mcp-server. Use uvx --from csv-mcp-server csv.

Verified recipe: load → stats → filter

# 1. Load
>>> {"jsonrpc":"2.0","id":10,"method":"tools/call","params":{"name":"load_csv","arguments":{"file_path":"/tmp/team.csv"}}}
<<< session_id: "e43447ef-...", shape: [5, 4], rows_affected: 5

# 2. Statistics
>>> {"jsonrpc":"2.0","id":11,"method":"tools/call","params":{"name":"get_statistics","arguments":{"session_id":"e43447ef-..."}}}
<<< experience_years: {count: 5, mean: 6.8, std: 3.42, min: 3, max: 12, sum: 34}

# 3. Filter (Python developers)
>>> {"jsonrpc":"2.0","id":12,"method":"tools/call","params":{"name":"filter_rows","arguments":{"session_id":"e43447ef-...","conditions":[{"column":"language","operator":"==","value":"Python"}]}}}
<<< rows_before: 5, rows_after: 2, rows_filtered: 3

All 3 calls returned success: true with correct data. The groupbyaggregate call after filter returned success: false — honest failure documented above.