> ## Documentation Index
> Fetch the complete documentation index at: https://www.landbase.com/docs/llms.txt
> Use this file to discover all available pages before exploring further.

# Continue a search or workflow

> Resume long-running MCP operations without starting duplicate work.

A search or workflow can continue after an MCP call returns or your client
disconnects. Keep its account ID, session ID, search handle, and run IDs in the
conversation. Use the identifier for the operation you already started.

## Choose the next call

| What you have | What to call |
| - | - |
| A running search `handle` | `get_search_result` with that handle |
| An exact `agent_run_id` | `get_agent_run` for one status read, or `wait_for_agent_run` for bounded polling |
| A `workflow_run_id` | `get_workflow_run` with that ID as `run_id` |
| An agent session ID after a lost trigger response | `list_agent_runs` or `get_latest_agent_run` with the original session and agent |

Pass the selected `account_id` explicitly on every status, polling, and
recovery call, including retries. The server does not remember account selection;
a handle or run ID does not replace it. For session recovery, have your assistant use the
original session and operation type. The [run recovery tool reference](https://www.landbase.com/docs/guides/mcp/mcp-supported-tools.md#list_agent_runs)
contains the required arguments.

Once you recover an exact run ID, prefer it to the latest-run lookup so a
later operation in the session cannot be mistaken for the original one.

## A wait expired

`wait_seconds` controls how long one tool call waits, up to 40 seconds.
A running result or `outcome: "wait_expired"` means you should continue with
the same handle or run ID. It does not mean the remote work stopped.

Ask your assistant:

> Resume the existing Landbase operation using the IDs already in this
> conversation. Check its status before starting anything new.

## A tool call timed out or the connection dropped

For a status read, retry the same read after any indicated retry delay. For a
call that starts work, first check whether it already started. Do not repeat
a search, paid workflow, upload, or publish call merely because its response
was lost.

If a search handle is available, poll it. If you have only its session ID,
recover the agent run and inspect any returned workflow ID. If no identifier
or reliable recovery result is available, report that the outcome is unknown
and ask before risking another operation. The server does not expose a general
recovery lookup for every type of write.

## Use the finished output

A search may return a `workflow_run_id` that still needs polling. Wait for
that workflow to reach `SUCCEEDED`. Use its `output_dataset_id` for the next
preview, workflow, or download; do not reuse the input dataset by accident.

A failed or cancelled result needs inspection. Do not automatically restart
it or assume that partial results form a finished output. Use
`get_dataset_lineage` when you need to trace the dataset and its processing.

See [Supported tools](https://www.landbase.com/docs/guides/mcp/mcp-supported-tools.md) for arguments and
[Limits and credits](https://www.landbase.com/docs/guides/mcp/limits-and-credits.md) for approval behavior.


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.