Loading ToolBench

Reading the current corpus…

langgraph-starter-kit MCP server · D · ToolBench

langgraph-starter-kit

github.com/ac12644/langgraph-starter-kit

Production-ready multi-agent starter kit with LangGraph — supports swarm, supervisor, HITL, structured output, and 6 LLM providers

2026-07-28+STREAMABLE_HTTP18 tools

53/100

Definition Quality50

Protocol Readiness65

Supportability50

Spec posture

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

Static source inference · medium confidence · evidence: stateless requests, Streamable HTTP

Current-spec patterns detected

Summary

18 tools with basic schemas and descriptions, but significant gaps in parameter documentation and error handling. Most tools have 1-3 parameters with type info, but lack depth in descriptions (many under 50 chars). No evidence of output schemas, pagination support, or recovery guidance. Tools like 'issue_refund' and 'delete_record' lack confirmation/dry-run patterns despite being irreversible. Naming is generally clear (verb_noun style), but descriptions are minimal and do not explain WHEN to use each tool or what happens on failure. No tool annotations (readOnlyHint, destructiveHint) visible in code.

Tools (18)

addread onlysource verified65/100

Add two numbers together

check_balanceread only50/100

Check customer account balance

create_ticketwrite50/100

Create a support ticket

delete_recorddestructivesource verified62/100

Delete a record by ID. Requires human approval.

echoread only50/100

Echo back the input text

escalate_to_humanwrite50/100

Escalate the issue to a human agent. Pauses the graph for approval.

initiate_returnirreversible50/100

Initiate a product return. Requires human approval.

issue_refundirreversible50/100

Issue a refund to a customer. Requires human approval.

list_recordsread only

Issues (5)

CRITICALconfirmation-request ↗

Irreversible operations (issue_refund, initiate_return, delete_record) lack confirmation/dry-run patterns. No evidence of approval gates or recovery guidance in descriptions.

issue_refundinitiate_returndelete_record

CRITICALtool ↗

No output schemas documented. Tools return unstructured responses; LLMs cannot plan downstream calls or extract required fields (e.g., customer_id, order_id for chaining).

lookup_customersearch_orderslookup_orderlist_ticketsretrieval

HIGHtool-description ↗

Descriptions are minimal (most under 50 chars) and lack context. No guidance on WHEN to use each tool, what happens on failure, or dependencies between tools.

echoweb_searchscrape_urlcheck_balancelist_records

HIGHrecovery-guide ↗

Recommendations

Add output schemas to all tools. Document returned fields (customer_id, order_id, ticket_id, etc.) so agents can chain calls without extra lookups.
Expand descriptions to 50-200 chars. Include WHEN to use (e.g., 'Call after lookup_customer fails to find by ID'), what happens on error, and any prerequisites.
Implement confirmation/dry-run for irreversible tools (issue_refund, delete_record, initiate_return). Return a confirmation token agents must pass to execute.
Add error classification to tool responses: retryable (network timeout), user-fixable (invalid customer_id), or fatal (permission denied). Include recovery hints.
Clarify transfer_to_* tool: rename to transfer_to_agent with explicit agent_name parameter and enum of valid targets. Document handoff protocol.

Score history

Overall score trend

First recorded score · v2 rubric

53/100

Scored	Grade	Overall	Spec posture	Rubric
2026-09-23	D	53	2026-07-28+	v2