mcpbench GitHub

client/resources-read

anthropic/claude-opus-4-8:medium 2025-11-25 docs: none score 100% musts 7/7 · shoulds 1/1 · mays 0/0 works yes · conformant yes

Checkpoints

MUST (any failure zeroes the score — 7/7 passed)

Checkpoint Description
✓ report-content Called report with the full text of the Deployment Runbook resource
✓ read-correct-uri Read the target resource via resources/read with params.uri
✓ pagination-followed Requested the second resources/list page using the issued nextCursor
✓ cursor-opaque-roundtrip Every sent cursor was a server-issued token echoed byte-for-byte
✓ session-header Echoed the issued Mcp-Session-Id on every subsequent request
✓ protocol-version-header Sent MCP-Protocol-Version on every post-initialize request
✓ accept-header Every accepted POST carried Accept: application/json, text/event-stream

SHOULD (the score fraction — 1/1 passed)

Checkpoint Description
✓ stop-at-end Stopped paginating after the response without a nextCursor

Usage & cost

Nominal cost
$0.338
Input tokens
36
Output tokens
8,030
Cache read
129,424
Duration
121s
Timestamp
2026-07-18 12:03:52Z
Bench version
0.3.0

Cost is nominal (public API pricing): token usage × published rates; actual marginal cost is subscription-covered.