GovTribe for Agents

Built for agent-assisted GovCon work

GovTribe gives agents structured tools, record context, and response references so they can help GovCon users research markets, qualify opportunities, organize capture work, and explain source data without guessing.

• Tool pages define available actions, valid arguments, and task-shaped examples for GovTribe AI and GovTribe MCP.
• Response pages describe payload fields, status objects, relationships, and interactive app outputs that agents can cite or pass to follow-on tools.
• Guides explain shared patterns for search construction, filters, context control, saved searches, and vector-store retrieval.
• Data model links connect tool output back to GovTribe record types, fields, identifiers, and lifecycle details.

Waiting for async research in GovTribe AI

When an async-capable model such as Astra has async tools enabled, its function schemas include task_handle and it receives the synchronous Wait_For_Tasks tool. These are GovTribe AI orchestration features, including for its MCP-backed function wrappers; they do not change public MCP schemas or add async support to hosted remote MCP tools. When these fields and tools are absent, use the ordinary synchronous workflow.

Give each async call a fresh, case-sensitive handle using 1–128 letters, digits, underscores, or hyphens. Handles remain reserved throughout the conversation, including after completion. Continue independent work, then call Wait_For_Tasks with a nonempty list of distinct handles whose results are needed. The wait yields until all selected tasks finish or fail while unrelated tasks continue. Results arrive on their original calls before the wait status. Do not use Shell sleeps to wait for application tools.

Wait status identifies each selected handle as completed, failed, or timed out. Each task has a ten-minute deadline from dispatch; waiting again does not extend it. Unknown handles and unfinished work from older turns return errors. Waiting again for a completed task returns its known status without executing it again or duplicating an already-delivered result. Report failures honestly rather than treating wait status as research evidence. Stop cancels waiting and unfinished work in the owning turn.