A capability-first MCP server toolkit for FastAPI with Streamable HTTP and stdio transport
PyMCP Kit presents a mixed quality picture. The server implements 22 tools across example and conformance test categories. Naming conventions are generally verb-based and clear (addNumbersTool, registerBonusTool, refreshWelcomeMemo), though inconsistent suffixing with 'Tool' and 'Memo' reduces clarity. Descriptions are present for all tools but vary significantly in depth, simple arithmetic tools have minimal 10-15 character descriptions that lack context for LLM selection, while notification and conformance tools provide better detail (60-120 chars). Input schemas are directly visible in the code with proper type annotations (float, string, int), but parameter descriptions are almost entirely absent from the schema definitions. The rubric requires descriptions for every parameter; none of the numeric parameters (a, b, length, width, n) have descriptions explaining their constraints, ranges, or purpose. Output schemas are not explicitly documented, tools return strings or untyped responses without structured field definitions. Error handling is minimal; most tools perform basic validation (e.g., promptEchoTool checks for 'crash' string) but provide no recovery guidance. The server demonstrates good composition, tools are single-purpose and chainable through RequestContext, but lacks the parameter descriptions and output documentation essential for production-grade definition quality.
Adds two numbers 'a' and 'b' and returns their sum.
Calculates the area of a rectangle given length and width.
Calls the /hello endpoint of the Flask API server running on port 5005 and returns the message.
Greets a person by their name.
Multiplies two numbers 'a' and 'b' and returns their product.
Echoes back the prompt provided, with input validation.
Notify subscribed clients that memo://welcome was updated.
Parameter descriptions are almost entirely absent from schema definitions. The rubric requires every parameter to have a description explaining its purpose, constraints, and range. Parameters like 'a', 'b', 'length', 'width', 'n', 'prompt', 'message', 'service' have type declarations but no descriptive text. This prevents LLMs from inferring valid ranges and proper parameter semantics.
Tool descriptions for arithmetic and simple utility tools are too brief (10-20 characters). Rubric baseline for tool descriptions is 194 chars average; descriptions under 20 chars cannot score above 40. These tools lack context on WHEN to use them, WHAT they return, and any prerequisites. Examples: 'Adds two numbers' (18 chars), 'Multiplies two numbers' (22 chars), 'Greets a person' (15 chars).
Inferred effective spec: 2026-07-28+.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | D | 51 | 2026-07-28+ | v2 |
| 2026-03-09 | F | 40 | 2024-11-05+ | v1 |
Register bonusTool at runtime and notify clients that the tools list changed.
Prompt template for release notes.
Send a structured log notification (notifications/message) to the connected client.
Return an audio content block.
Request user input from the client.
Request elicitation with schema default values.
Request elicitation exercising all enum variants.
Return an embedded resource content block.
Return a tool error result.
Return an image content block.
Return mixed text, image, and resource content.
Request LLM sampling from the client.
Return a simple text content block.
Emit log notifications during execution.
Send progress notifications during execution.
Output schemas are not documented. Tools return strings or untyped responses without explicit field definitions. Rubric requires 'Document the output schema. LLMs need to know what fields to expect so they can plan downstream tool calls.' Example: addNumbersTool returns a string like 'Sum of 2 + 3 = 5' but callers cannot programmatically extract the numeric result.
Inconsistent naming conventions reduce clarity. Tool names mix suffixes ('Tool' appended to arithmetic/utility tools but not to resource/prompt handlers). Naming pattern test_* for conformance tools is generic; does not follow verb_noun convention (e.g., 'assert_text_rendering', 'verify_image_output' would be clearer). Rubric: 'The tool name alone should convey what happens when called.'
Error handling is minimal and lacks recovery guidance. promptEchoTool validates input with a hardcoded 'crash' check, but the error message 'Invalid input. Please try again.' does not tell the LLM what to do next or which constraints were violated. Rubric: 'Error responses must tell the LLM what to do next: ... A raw error code or stack trace gives the agent nothing to act on.'