everyrowby futuresearch
by futuresearch

Progress Monitoring

everyrow operations typically take 1–10+ minutes depending on dataset size and operation type. Both the Python SDK and the MCP tools provide progress monitoring and a session URL where you can watch tables update in real-time.

What to Expect

Web UI

Every operation is part of a session, which you can view at https://everyrow.io/sessions/<session_id>. You can open it in your browser to see:

  • Real-time progress for each row
  • Web searches each agent ran and the pages it read
  • Explanation for each result, including links to sources
  • Data visualizations

Python SDK

To see progress updates while a task is running, use the print_progress callback:

from everyrow import print_progress

result = await task.await_result(on_progress=print_progress)

Output:

 0/10   0% | 10 running
 2/10  20% |  8 running
 4/10  40% |  6 running
 6/10  60% |  4 running
 8/10  80% |  2 running
10/10 100%

You can also provide a custom on_progress callback for programmatic progress handling (see below).

MCP Tools

When you run an everyrow operation via MCP:

  1. The tool returns immediately with a session URL
  2. Progress updates appear every few seconds during execution
  3. Results are saved as a CSV file when the operation completes
  4. If you've installed the plugin, a desktop notification (macOS and Linux) tells you when it's done

The workflow:

everyrow_agent        →  start the operation, get a task_id and session URL
everyrow_progress     →  check status (blocks for a few seconds, then returns progress)
everyrow_progress     →  check again (the agent loops automatically)
everyrow_results      →  download results when complete

The agents handle the polling loop automatically. You'll see progress in the conversation like:

Running: 20/50 complete, 30 running (45s elapsed)

And when it finishes:

Completed: 50/50 (0 failed) in 100s
...
Saved 50 rows to /path/to/output.csv

Claude Code Integration

Progress Bar

For a persistent progress bar in the Claude Code terminal footer, add this to either your global settings (.claude/settings.json) or project settings (.claude/project-settings.json), depending on where you have the MCP server configured:

{
  "statusLine": {
    "type": "command",
    "command": "<path-to-plugin>/everyrow-mcp/scripts/everyrow-statusline.sh",
    "padding": 1
  }
}

This shows a live progress bar:

everyrow ████████░░░░░░░ 42/100 23s   view

The "view" link is clickable in terminals that support hyperlinks (iTerm2, kitty, WezTerm, Windows Terminal) and opens the session dashboard in your browser.

The status line and hook scripts require jq for JSON parsing:

# macOS
brew install jq

# Linux
apt install jq

Note: After adding the config, restart Claude Code. Status line settings are loaded at startup only.

Stop Guard

The plugin includes a stop guard hook that prevents Claude from ending its turn while an operation is running. If you see:

Stop hook error: [everyrow] Task abc123 still running. Call everyrow_progress(task_id="abc123") to check status.

This is expected behavior, not an error. The "error:" prefix is a known cosmetic issue in Claude Code. The hook is working correctly — it's keeping the agent focused on your running operation.

Python SDK Progress

For printing progress updates as a task runs, use the provided print_progress callback:

from everyrow import print_progress

result = await task.await_result(on_progress=print_progress)

Or provide a custom callback:

from everyrow.generated.models import TaskProgressInfo

def my_progress_handler(progress: TaskProgressInfo):
    print(f"{progress.completed}/{progress.total} done, {progress.failed} failed")

result = await task.await_result(on_progress=my_progress_handler)

The callback receives a TaskProgressInfo object and only fires when the progress snapshot changes.