mcpbench GitHub

Scenarios

Each scenario runs per (spec version × docs condition). Client scenarios: the agent must call an instrumented reference server conformantly. Server scenarios: the agent builds a server that survives our conformance probes. Checkpoints are asserted from runner-side observation only.

client/bad-version-response

client 2025-11-25

client/composite

client 2025-11-25

client/discover-instructions

client 2026-07-28

client/drop-stream-mid-call

client 2025-11-25

client/elicit-form

client 2025-11-25

client/elicit-url

client 2025-11-25

client/ema-id-jag-flow

client 2025-11-252026-07-28

client/get-stream

client 2025-11-25

client/header-mirroring

client 2026-07-28

client/initialize-handshake

client 2025-11-25

client/mrtr-basic

client 2026-07-28

client/mrtr-elicit-url

client 2026-07-28

client/mrtr-multi-round

client 2026-07-28

client/mrtr-no-state

client 2026-07-28

client/no-batching

client 2025-11-25

client/oauth-cimd

client 2025-11-252026-07-28

client/oauth-cimd-over-dcr

client 2025-11-252026-07-28

client/oauth-dcr-application-type

client 2026-07-28

client/oauth-iss-mismatch

client 2026-07-28

client/oauth-pkce

client 2025-11-252026-07-28

client/oauth-refresh-token

client 2026-07-28

client/param-headers

client 2026-07-28

client/pkce-unsupported-as

client 2025-11-252026-07-28

client/prompts-get

client 2025-11-25

client/protocol-version-header

client 2025-11-25

client/resources-read

client 2025-11-25

client/session-echo

client 2025-11-25

client/session-recovery

client 2025-11-25

client/sse-response

client 2025-11-25

client/stalled-tool

client 2025-11-25

client/stateless-composite

client 2026-07-28

client/stateless-prompts

client 2026-07-28

client/stateless-resources

client 2026-07-28

client/stateless-sse-response

client 2026-07-28

client/stateless-tool-call

client 2026-07-28

client/task-cancel

client 2026-07-28

client/task-polling

client 2026-07-28

client/tool-call-basic

client 2025-11-25