Loading ToolBench

Reading the current corpus…

mcpg MCP server · C · ToolBench

mcpg

A governed MCP endpoint gateway that acts as a reverse proxy for MCP servers, providing rate limiting, policy enforcement, authorization, audit logging, and multi-protocol support (HTTP, stdio). Supports both legacy (2025-11-25) and modern (2026-07-28) MCP protocol versions.

2026-07-28+STREAMABLE_HTTP2 tools

63/100

Definition Quality62

Protocol Readiness78

Supportability56

Spec posture

Inferred effective spec 2026-07-28+·current baseline 2026-07-28

Static source inference · medium confidence · detected: Logging, Dynamic Client Registration

Deprecated protocol patterns detected

Summary

MCPG exposes 2 tools with explicit schemas and descriptions. Both tools have clear verb-noun naming (tools/list, tools/call) and detailed descriptions (150+ chars each). Input schemas are present with typed parameters and descriptions. However, output schemas are not documented in the visible code, and error handling guidance is absent. The tools reference advanced MCP patterns (MRTR, idempotency, progress tokens, task support) but lack descriptions of what those features mean for LLM behavior. No tool annotations (readOnlyHint/destructiveHint) are visible despite tools/call being destructive.

Tools (2)

tools/callwriteauth50/100

Dispatch tools/call on the modern wire. Executes a tool with the given name and arguments. Supports task-augmented execution (SEP-2663), MRTR resumption (SEP-2322), idempotency (SEP-2133), progress tokens (SEP-2575), and per-request log-level control. Enforces taskSupport constraints and policy gates.

tools/listread only50/100

Dispatch tools/list on the modern wire. Enumerates available tools with visibility filtering, catalog pagination, and caching metadata. Returns a paginated list of tool descriptors with cache scope and TTL.

Issues (5)

HIGHtool ↗

Output schemas not documented. Neither tool specifies what fields are returned, their types, or structure. LLMs cannot plan downstream calls or extract required data without documented return types.

tools/listtools/call

HIGHtool-annotations ↗

tools/call is destructive (WRITE risk) but lacks tool annotations. No readOnlyHint or destructiveHint visible. LLMs cannot distinguish safe vs unsafe tools without explicit annotations.

tools/call

HIGHrecovery-guide ↗

Error handling guidance absent. Descriptions reference advanced features (MRTR, idempotency, task support) but do not explain failure modes or recovery paths. LLMs lack actionable error context.

tools/listtools/call

MEDIUMpaginated-result ↗

tools/list pagination cursor is opaque. Description says 'Opaque pagination cursor' but does not explain when to stop iterating, what happens on invalid cursor, or whether results are ordered. Pagination guidance incomplete.

Recommendations

Document output schemas for both tools. Specify: tools/list returns {tools: [{name, description, inputSchema, ...}], nextCursor?, totalCount?}; tools/call returns {content: [{type, text|toolUse}], isError, ...}.
Add tool annotations: mark tools/call with destructiveHint=true and idempotentHint=true (if idempotencyKey is provided). Mark tools/list with readOnlyHint=true.
Expand error handling in descriptions. E.g., 'If tool not found, returns error code TOOL_NOT_FOUND with available tool names. If arguments invalid, returns INVALID_ARGUMENT with schema violations. Retry-safe: use idempotencyKey.'
Clarify pagination contract: 'Cursor is opaque. Omit cursor on first call. If nextCursor is null, no more pages. Results ordered by tool name. Invalid cursor returns error INVALID_CURSOR.'
Document meta parameter constraints: 'progressToken: optional UUID for progress updates. requestLogLevel: optional, one of debug|info|warn|error. idempotencyKey: optional UUID, enables safe retries. MRTR fields: internal, do not set manually.'

Score history

Overall score trend

First recorded score · v2 rubric

63/100

Scored	Grade	Overall	Spec posture	Rubric
2026-09-23	C	63	2026-07-28+	v2