omniseek_gather
Run multiple read-only tools in parallel and collect results in one response, simplifying multi-source queries.
Instructions
Run N independent read-only eye tools IN PARALLEL, returning results in one response.
Fully-qualified MCP name: mcp__omniseek__omniseek_gather (server name is omniseek; there is no omniseek-eye server).
The agent decides WHAT to call (judgment). OmniSeek executes them (mechanical). Each call runs independently; one failure does not affect others. Calls that depend on a prior call's result belong in a SEPARATE gather (the agent reads this batch first, then decides the next batch).
calls: [{"tool": "omniseek_search", "args": {"query": "..."}}, ...]
Bounded: max 10 calls. Read-only tools only.
Calls naming search-index venues / nowcoder serialize through one paced backend (see server
instructions (10)); >2 per gather adds a warnings entry.
wait_s: the patience budget. gather returns when all calls finish OR wait_s elapses,
whichever comes first; calls still running are reported with status "warming" (their
background threads keep going and warm the cache — pick them up later with
staleness="cache_only" or a second gather).
Returns: {results: [{index, tool, status, result|error|hint}, ...],
elapsed_s, completed, warming, failed, total}
On an errored call whose failure is a call-signature mismatch (a wrong / missing argument), hint
names the tool's REAL parameters (e.g. "omniseek_read takes: target, start_char, max_chars, ...").
Input Schema
| Name | Required | Description | Default |
|---|---|---|---|
| calls | Yes | ||
| wait_s | No |