fix: handle Pydantic TextContent objects in MCP response parser

[bokelley]

Summary

Fixes AttributeError in v1.0.3 where response parser couldn't handle MCP SDK's Pydantic TextContent objects, and refactors the architecture to establish proper separation of concerns.

Problem

The MCP SDK returns Pydantic objects (e.g., TextContent), not plain dicts. In v1.0.3, response parsing logic was added that used item.get("type") and item.get("text"), which only works on dicts:

AttributeError: 'TextContent' object has no attribute 'get'

This was a regression - v1.0.2 worked correctly.

Root Cause

The MCP adapter was passing MCP SDK-specific Pydantic objects directly to the generic response parser, violating protocol boundary separation. The parser had to handle both parsing AND protocol translation.

Solution

Architectural fix with proper separation of concerns:

1. MCP Adapter (mcp.py): Added _serialize_mcp_content() to convert Pydantic objects to plain dicts at the protocol boundary
2. Response Parser (response_parser.py): Simplified to only handle plain dicts, removed dual-mode helper function
3. Tests: Added serialization tests to protocol adapter, removed Pydantic mocks from parser tests

Benefits

• ✅ Clean separation: adapters translate, parsers parse
• ✅ Simpler type signatures: list[dict[str, Any]] (no more | Any)
• ✅ Parser tests use plain dicts (no MCP SDK dependency)
• ✅ More obvious code (removed clever helper function)
• ✅ Follows adapter pattern correctly
• ✅ Establishes pattern for future MCP content types (images, resources)

Commits

1. Initial fix: Handle Pydantic objects with get_field() helper (f29179b)
2. Architectural refactor: Move serialization to protocol boundary (7d2a521)

Tests

Response Parser Tests:

• ✅ Parsing dict-based MCP text content
• ✅ Multiple content items
• ✅ Empty content error handling
• ✅ Invalid schema error handling
• ✅ Empty text content skipping

Protocol Adapter Tests:

• ✅ Serializing dict content (pass-through)
• ✅ Serializing Pydantic v2 objects (model_dump)
• ✅ Serializing mixed content (dicts + Pydantic)

All tests passing (15/15 for affected modules).

Test plan

• Run unit tests for response_parser (all passing)
• Run unit tests for protocol adapters (all passing)
• Verify serialization handles Pydantic v2 model_dump()
• Verify backward compatibility with dict-based tests
• Test with real MCP agent to verify fix works in production

🤖 Generated with Claude Code