mcpbench GitHub

client/stateless-prompts

moonshotai/kimi-k2.7-code 2026-07-28 docs: full score 0% task-failure musts 7/8 · shoulds 2/2 · mays 0/0 works no · conformant no

Checkpoints

MUST (any failure zeroes the score — 7/8 passed)

Checkpoint Description Detail
report-content Called report with the exact prompt message text tools/call for "report" was never sent
prompts-get-args Sent prompts/get with name release_notes and arguments {version: "2.4.1"}
prompts-list-first A successful prompts/list preceded the successful prompts/get
stateless-meta Every request carried _meta protocolVersion and clientCapabilities
mcp-method-header Mcp-Method header mirrored the JSON-RPC method on every request
mcp-name-header-prompt Mcp-Name header mirrored params.name on prompts/get requests
version-header-match MCP-Protocol-Version header present and matching the _meta value
accept-header Every accepted POST carried Accept: application/json, text/event-stream

SHOULD (the score fraction — 2/2 passed)

Checkpoint Description
clientinfo-every-request Included clientInfo in _meta on every request
no-legacy-lifecycle Never sent a legacy initialize request or an Mcp-Session-Id header

Usage & cost

Nominal cost
$0.072
Input tokens
41,084
Output tokens
3,007
Cache read
262,720
Duration
86s
Timestamp
2026-07-18 16:52:55Z
Bench version
0.3.0

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