Unstable preview. Pages, data and API fields can change or break without notice, so please don't rely on this for anything critical yet. See what has changed.
Agent Service Directory
Menu

A2AWire Benchmark: Cold-Walk Capitals Check

MCP server · a2awire.com · found in a public list · checked since 2026-11-08

Live, handshake completed · latest check

As JSON

Runs an agent benchmark on capital-city facts from a cold-walk dataset.

Our summary of what its own text says it does: automated, not reviewed by a person (placement-ai/7). What this means

Placed in Evaluate AI agents and prompts: Run prompts, agents or tool calls against test cases and return scores, regressions and failures (checks of other people's services: identity/security/agent-services).

https://a2awire.com/mcp/benchmarks/cw-capitals-1108-a17627f5/http MCP endpoint (Streamable HTTP)

The name is the registry's label, not ours. Not reviewed by a person. See 135 endpoints from a2awire.com.

Our checks

An unpaid request daily, from eastern North America

Last 2 checks, oldest to newest; room for 28 more

2 passed, 0 failed, 28 not run.

Passed, 30 daysshort history2 of 2 run
Checks missed0in 30 days
Response, latest575 mswhole handshake
Scheduledailyset by usage; rule below
Checked since2026-11-081 day
Freshnessfreshagainst its schedule

Checked daily. Daily: 5 to 99 payers, and every new service until it has been probed 3 times.

Response times

WindowChecksTimeoutsMedian90th95th99thFastestSlowest
24 hours10——————
7 dayssince 2026-11-0820——————
30 dayssince 2026-11-0820——————

Milliseconds, from eastern North America, timing the whole MCP handshake (initialize to the tools/list answer). Timeouts count at the 10000 ms limit. A dash means too few checks for that figure. As of . Definitions and method.

Details

Tools listed by the server

a2awire_guidebenchmark_corpusbenchmark_finalize_runbenchmark_get_resultsbenchmark_start_runbenchmark_submit_answersbenchmarks_getbenchmarks_listbirthcheck_earningsconfirm_keys_persisteddata_session_attach_escrowdata_session_funddata_session_funding_packagedata_session_opendata_session_querydiscover_agentsfind_paid_workget_agent_contractget_recommended_actionhire_and_executeonboard_startregisterrequest_testnet_usdcverify_contract

Tools
25
This list since
2026-11-08
First listed to us
2026-11-08
List changes, 90 days
0

Names as the server reported them at our latest completed handshake (mcp-handshake/2); a listed tool is not a tested one. Definitions

Access

Our latest check
Nothing asked
Stated
not stated
Payment model
not known
Currency
not known

"Nothing asked" describes one plain request, not every operation. Statements are claims, not our measurements. Definitions

Interface it declares

Protocol
MCP 2025-06-18
Capabilities
promptsresourcestools
Transports
streamable-http

As its initialize reply stated it at our check on (mcp-handshake/2): a claim, not tested. Definitions

Calls it declares

  • benchmarks_list
  • benchmarks_get slug* agent_id
  • benchmark_corpus slug agent_id
  • benchmark_start_run slug agent_id
  • benchmark_submit_answers run_id* answers* agent_id
  • benchmark_finalize_run run_id* agent_id
  • benchmark_get_results run_id* agent_id
  • request_testnet_usdc address reason
  • confirm_keys_persisted → keys_persisted
  • data_session_open listing_id buyer_address max_queries proof_escrow_id open_tx_hash listing_slug
  • data_session_funding_package session_id*
  • data_session_fund session_id*

From the tool list it returned to our check (request-shape/2). * marks a required parameter; after → the fields it says it returns. What it declares, not that a call works.

Listing

In the capability tree: Compute, code and storage / Developer data / Validate and test / Evaluate AI agents and prompts (primary). Placed by three checks by a small AI model: placed, found blind from its tasks, and verified against its own text (placement-ai/7, tree/3) on 2026-11-08. Wrong place? Tell us.

In the directory: meets our listing criteria (listing-criteria/6): ✓ says what it does; ✓ its latest check passed.

Recent probes

WhenResultTime to headers

cloudflare:EWR		 Live, handshake completed HTTP 200 575 ms

cloudflare:EWR		 Live, handshake completed HTTP 200 237 ms

What these figures do not show

  • A passing check means the server completed the MCP handshake with us (initialize, then tools/list, with no credentials; mcp-handshake/2) or asked for credentials first (HTTP 401). It does not show that the server's tools work or what they return. Tool names are as the server listed them.
  • Listed in the Official MCP Registry; this data comes from the Official MCP Registry. Its listing's description is not republished.
  • Limited evidence: monitored for 1 day (less than 30).

Something wrong here?

Operate this service and want it reviewed, corrected or removed? Email hello@agentservicedirectory.com or use the opt-out page. A wrong place in the tree? Tell us.

Machine-readable: /api/v1/services/cand_30082ca19430184b