:bug: git issue always wait for timeout to drop an mcp call

[davidhadas]

Describe the bug

(1)
If the git issue agent get an error during an mcp call, it doe snot handle it as an error instead it continue to work on the problem until it timeout

(2)
Some requests lead the agent to take forever due to unbounded mcp input resulting in oversized llm request and resulting error

Steps To Reproduce

(1)
in HITL refuse to authenticate and than do second task to the same agent - agent wait for 10 min before it serve the new task
(2) ask many issues

Expected Behavior

(1)
An error arriving during connect should be treated as error immediately.

(2)

• Bound the fetch / slice results. Constrain the MCP tool args (per_page/limit) and/or slice the observation to top-N before it ever reaches the prompt. Prevents the overflow at the source — which is the only way to reliably kill the CrewAI summarize loop from our side (addresses causes 1, 2, and indirectly 3+4).
• Cap tool output size feeding the model (token/char budget), with a user-facing "results truncated to N" note (cause 6).
• Honor cancellation. Wire A2A cancel/disconnect into the run: run the crew as a cancellable task and cancel it on disconnect; implement cancel() instead of the raise. Note CrewAI is largely synchronous under kickoff_async, so true mid-flight abort may require running it in a cancellable executor and checking a stop flag between steps — needs verification of how much CrewAI cooperates.

Kagenti Version

No response

Additional Context

No response