Loading ToolBench

Reading the current corpus…

mcp-router MCP server · F · ToolBench

mcp-router

An MCP router/gateway that manages multiple upstream MCP connectors (HTTP and stdio), provides policy-based tool access control, handles OAuth authentication, and exposes a unified tool catalog to MCP clients

2026-07-28+STREAMABLE_HTTP7 tools

49/100

Definition Quality50

Protocol Readiness68

Supportability36

Spec posture

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

Static source inference · medium confidence · evidence: Streamable HTTP

Current-spec patterns detected

Summary

mcp-gateway exposes 7 tools with basic descriptions but critical gaps in schema completeness and parameter documentation. Tools 1-7 all have descriptions (10-80 chars, mostly under the 194-char baseline), but input schemas are severely incomplete: tools/call and prompts/get show minimal schema detail (only 'name' and 'arguments' with generic descriptions); resources/read has only 'uri' with no type info visible; initialize, tools/list, resources/list, prompts/list lack visible input schemas entirely. No parameter descriptions beyond the tool-level summary. Output schemas are undocumented. Error handling is absent, no recovery guidance, no categorization. Naming is acceptable (verb_noun pattern: list, read, call, get) but generic. No tool annotations (readOnlyHint, destructiveHint, idempotentHint) despite risk labels in metadata. This is a gateway/router pattern, not a direct tool provider, which limits direct schema visibility, but the HTTP adapter should still expose full schemas to clients.

Tools (7)

initializeread onlyauthsource verified43/100

Initialize MCP session with the gateway, returning protocol version, capabilities, and server info

prompts/getread onlyauthsource verified57/100

Get a prompt from a connected connector with optional arguments, enforcing policy-based access control

prompts/listread onlyauthsource verified47/100

List all available prompts from connected connectors that are allowed by the client's policy

resources/listread onlyauthsource verified47/100

List all available resources from connected connectors that are allowed by the client's policy

resources/readread onlyauthsource verified55/100

Read a resource from a connected connector, enforcing policy-based access control

tools/callwriteauthsource verified57/100

Call a tool on a connected connector with the given arguments, enforcing policy-based access control

tools/list

Issues (5)

CRITICALconstrained-input ↗

Input schemas missing or incomplete for 5 of 7 tools (initialize, tools/list, resources/list, prompts/list, and partial for tools/call, resources/read, prompts/get). Only 'name'/'uri'/'arguments' visible with no type constraints, enums, or parameter descriptions.

initializetools/listresources/listprompts/listtools/callresources/readprompts/get

CRITICALtool ↗

No output schemas documented for any tool. Clients cannot infer response structure, forcing them to guess field names and types. Breaks tool chaining.

initializetools/listtools/callresources/listresources/readprompts/listprompts/get

HIGHrecovery-guide ↗

No error handling guidance. Tools lack recovery hints (e.g., 'Tool not found. Call tools/list to discover available tools'). Errors are not categorized as retryable, user-fixable, or fatal.

tools/callresources/read

Recommendations

Add full JSON Schema input definitions for all 7 tools. For tools/call and prompts/get, document that 'arguments' must match the schema of the target tool/prompt (discovered via tools/list or prompts/list). Include type, required fields, and constraints.
Document output schemas for each tool. E.g., tools/list should return {tools: [{name, description, inputSchema, isReadOnly}], ...}. resources/read should return {uri, mimeType, contents}. This enables LLM planning.
Add parameter descriptions for 'name', 'uri', 'arguments'. E.g., 'name: Tool identifier in format connector__toolname (e.g., slack__send_message). Call tools/list to discover available tools.'
Implement error recovery guidance. E.g., tools/call error: 'Tool "slack__unknown" not found. Available tools: [list]. Call tools/list to refresh.' Categorize errors as retryable (network timeout) vs user-fixable (invalid tool name).
Expose tool annotations via MCP protocol. Map risk labels (READ_ONLY, WRITE) to readOnlyHint and destructiveHint in tool definitions. This helps LLMs reason about side effects and retry safety.

Score history

Overall score trend

First recorded score · v2 rubric

49/100

Scored	Grade	Overall	Spec posture	Rubric
2026-09-23	F	49	2026-07-28+	v2