- Add pydantic_core.to_json() serialization for non-string prompt arguments
- Update type annotations to accept dict[str, Any] instead of dict[str, str]
- Add focused tests covering specific scenarios:
* Client always serializes non-string args regardless of server types
* Integration with server-side type conversion
* Client serialization error with specific PydanticSerializationError
* Server deserialization error with specific McpError match
This ensures MCP protocol compliance while maintaining developer experience
with typed arguments.
🤖 Generated with [Claude Code](https://claude.ai/code)
Co-Authored-By: Claude <noreply@anthropic.com>
Add proper null checks and type assertions for prompt argument
handling in tests to satisfy pyright strict typing.
🤖 Generated with [Claude Code](https://claude.ai/code)
Co-Authored-By: Claude <noreply@anthropic.com>
Change from 'Arguments must be strings conforming to this JSON schema'
to 'Provide as a JSON string matching the following schema' for clearer
instruction to LLMs about string format requirements.
🤖 Generated with [Claude Code](https://claude.ai/code)
Co-Authored-By: Claude <noreply@anthropic.com>
- Fix ValueError -> PromptError for consistent error handling
- Add automatic JSON schema descriptions to non-string prompt arguments
- Include comprehensive tests for argument description enhancement
- Verify enhanced descriptions are visible via MCP protocol
This helps developers understand the expected string format for complex
types when calling prompts from MCP clients.
🤖 Generated with [Claude Code](https://claude.ai/code)
Co-Authored-By: Claude <noreply@anthropic.com>
Updated FunctionPrompt.render() to accept dict[str, Any] instead of
dict[str, str | Context] to preserve the developer experience of
passing properly typed arguments while also supporting string-only
arguments from MCP clients.
The _convert_string_arguments method now intelligently handles both
scenarios:
- Already-typed arguments are passed through unchanged
- String arguments are converted to expected types when needed
This maintains backward compatibility while enabling MCP spec compliance.
🤖 Generated with [Claude Code](https://claude.ai/code)
Co-Authored-By: Claude <noreply@anthropic.com>
- Add comprehensive server inspection utility supporting both FastMCP 1.x and 2.x
- Create detailed info dataclasses for tools, prompts, resources, and templates
- Implement CLI command with path:object notation and JSON output
- Add version reporting (fastmcp_version, mcp_version, server_version)
- Include comprehensive unit tests for utilities and CLI
🤖 Generated with [Claude Code](https://claude.ai/code)
Co-Authored-By: Claude <noreply@anthropic.com>
Resolves 307 redirect from /mcp to /mcp/ by replacing Mount with explicit Route entries for both path variations. The MCP specification requires a single endpoint to handle both GET and POST requests without redirects.
Changes:
- Replace Mount with Route for both /mcp and /mcp/ paths
- Add PathNormalizingASGIApp wrapper to normalize trailing slashes for MCP SDK
- Apply same pattern to SSE transport for consistency
- Ensure both auth and non-auth configurations work correctly
This fixes the redirect issue mentioned in GitHub issue #828.
🤖 Generated with [Claude Code](https://claude.ai/code)
Co-Authored-By: Claude <noreply@anthropic.com>
The previous example only allowed GET methods, causing browser preflight
requests to fail with "Disallowed CORS method" errors. Updated to include
the required parameters for proper CORS support with MCP clients.
🤖 Generated with [Claude Code](https://claude.ai/code)
Co-Authored-By: Claude <noreply@anthropic.com>
The audience parameter was typed as `str | None` but the implementation
already supported `List[str]`. This fix aligns the type annotations with
the actual functionality and adds comprehensive validation logic for all
audience type combinations.
🤖 Generated with [Claude Code](https://claude.ai/code)
Co-Authored-By: Claude <noreply@anthropic.com>
