Python: Persist hosted MCP call/results as canonical mcp_call output

[Hameedkunkanoor]

Motivation and Context

This PR recreates and carries forward the original hosted MCP persistence fix from #5950, adapted on top of current main and finalized with follow-up CI/type-safety hardening.
Fix for #5546.

Problem addressed:

• Follow-up turns could fail with a 400 when replayed hosted MCP history contained an unbalanced tool-output shape.
• Hosted MCP call and result could be persisted in split form, which could replay as orphaned output.

Scenario supported:

• Cross-turn hosted MCP conversations where tool call identity and tool result remain paired in persisted/replayed history.

Dependency and rollout note:

• This change depends on SDK APIs in azure-ai-agentserver-responses and now targets azure-ai-agentserver-responses>=1.0.0b7,<2.
• This PR is the functional continuation of Python: Persist hosted MCP call/results as canonical mcp_call output #5950 with dependency floor updated from b6 to b7.

Description

Overall approach:

• Keep foundry_hosting write-side persistence on canonical single-item mcp_call representation and keep replay reconstruction aligned with that shape.

Changes made:

1. Preserve original MCP call id when opening MCP builders via item_id/call_id mapping.
2. Streaming conversion path: mcp_server_tool_result completes the active mcp_call builder instead of falling back to custom_tool_call_output.
3. Non-streaming conversion path: adjacent mcp_server_tool_call + mcp_server_tool_result are coalesced into one completed mcp_call output item.
4. Replay reconstruction: persisted mcp_call items with output reconstruct back into MCP call/result content.
5. Dependency update: bump foundry_hosting floor from azure-ai-agentserver-responses>=1.0.0b5,<2 to >=1.0.0b7,<2.
6. Test coverage: scenario and regression tests for streaming/non-streaming persistence, replay reconstruction for both item shapes, and multi-turn round-trip behavior.
7. Follow-up hardening from recreation cycle: fixes for missing Mapping import, pyright/mypy typing issues in MCP output stringification, and safer mapping serialization fallback.

Files changed:

• python/packages/foundry_hosting/agent_framework_foundry_hosting/_responses.py
• python/packages/foundry_hosting/tests/test_responses.py
• python/packages/foundry_hosting/pyproject.toml

Validation

• Updated unit tests in tests/test_responses.py for MCP call/result coalescing, call-id mapping, output persistence, and replay reconstruction.
• CI follow-up fixes were applied to satisfy package checks (typing/lint) on the recreated PR branch.

Risk / Impact

• Low-to-moderate behavioral risk in response-event shaping for hosted MCP events.
• Expected impact: deterministic canonical MCP output structure for clients and replay paths, avoiding orphaned function/tool output in follow-up turns.

Relationship to Prior PR

• Python: Persist hosted MCP call/results as canonical mcp_call output #6070 is a recreation/continuation of Python: Persist hosted MCP call/results as canonical mcp_call output #5950 on a newer mainline state, with equivalent intent and behavior plus follow-up CI/type robustness fixes and the b7 dependency floor.

Contribution Checklist

• The code builds clean without any errors or warnings
• The PR follows the Contribution Guidelines
• Unit tests were updated for the new behavior
• No breaking API surface change intended

[Copilot]

Pull request overview

Note

Copilot was unable to run its full agentic suite in this review.

This PR updates the Foundry hosting responses adapter to coalesce hosted MCP tool call + tool result into a single mcp_call output item (including output), adds round-trip coverage for the new behavior, and bumps the azure-ai-agentserver-responses dependency to pick up the needed model/events support.

Changes:

• Coalesce hosted MCP mcp_server_tool_call + mcp_server_tool_result into a single mcp_call output item (non-streaming and streaming).
• Reconstruct MCP result content when reading mcp_call items that include output.
• Add tests for persistence, streaming emission, reconstruction, and multi-turn history replay; bump azure-ai-agentserver-responses to b6.

Reviewed changes

Copilot reviewed 3 out of 3 changed files in this pull request and generated 2 comments.

File	Description
python/packages/foundry_hosting/agent_framework_foundry_hosting/_responses.py	Adds hosted-MCP coalescing and output stringification; updates streaming handler and item-to-message reconstruction to carry MCP output.
python/packages/foundry_hosting/tests/test_responses.py	Adds regression tests ensuring MCP calls/results persist/stream as a single mcp_call and replay correctly across turns.
python/packages/foundry_hosting/pyproject.toml	Bumps azure-ai-agentserver-responses to a newer beta required for MCP output support.

[moonbox3]

Python Test Coverage Report •

File	Stmts	Miss	Cover	Missing
packages/foundry_hosting/agent_framework_foundry_hosting
_responses.py	747	120	83%	183–186, 251, 328–329, 339, 376, 431, 445, 495, 498–502, 521, 524, 530, 532, 553–555, 584–586, 591, 593, 600, 602–603, 605, 607, 613, 617, 619–621, 625, 628, 633–639, 642–643, 645–646, 654–659, 960, 973, 1442–1444, 1446, 1493–1494, 1496–1497, 1499–1500, 1502–1503, 1508, 1517, 1520–1522, 1524, 1538, 1551, 1596–1597, 1599, 1604–1608, 1610, 1617–1618, 1620–1621, 1627, 1629–1633, 1640, 1646, 1668, 1674, 1680, 1682, 1684–1691, 1699, 1701, 1745–1747, 1757–1758
TOTAL	36359	4324	88%

Python Unit Test Overview

Tests	Skipped	Failures	Errors	Time
7245	34 💤	0 ❌	0 🔥	1m 53s ⏱️