Sato Hub
Daily checks

Reliability Index

See which crypto agent tools answered our daily check, ranked by success rate.

See the rankingGet the data (opens in a new tab)
  • Updated2026-10-22
  • Minimum record14 days
RankedRankedA tool needs at least 14 days of daily checks to be ranked. A perfect record over six days is not a record.
432
tools with 14+ days of checksas of
Answered every check
333
of the ranked toolsas of
Under 80%Under 80%Counts tools whose misses we can place. A tool whose every miss has an unknown cause is shown, ranked last, and left out of this number.
45
of ranked tools with a known causeas of
MCP-checkedMCP-checkedThe strongest check we run: we start an MCP session and ask the tool to list what it offers. A parked domain cannot fake that.
78
answer a real MCP handshakeas of

The ranking

28 tools
All toolsInstall reproduced

AnsweredNo answerNo check counted

  1. Almanak
    Agent FrameworkHTTP onlyInstall reproduced

    100%

    answered

    2026-09-2329 answered0 no answer1 no check2026-10-22

    90days on record

  2. Arbitrum Vibekit
    Agent FrameworkHTTP only

    100%

    answered

    2026-09-2329 answered0 no answer1 no check2026-10-22

    90days on record

  3. Claude Code
    Agent FrameworkHTTP onlyInstall reproduced

    100%

    answered

    2026-09-2329 answered0 no answer1 no check2026-10-22

    90days on record

  4. ElizaOS
    Agent FrameworkHTTP onlyInstall reproduced

    100%

    answered

    2026-09-2329 answered0 no answer1 no check2026-10-22

    90days on record

  5. Fetch.ai uAgents
    Agent FrameworkHTTP onlyInstall reproduced

    100%

    answered

    2026-09-2329 answered0 no answer1 no check2026-10-22

    90days on record

  6. GAME SDK (Virtuals Protocol)
    Agent FrameworkHTTP onlyInstall reproduced

    100%

    answered

    2026-09-2329 answered0 no answer1 no check2026-10-22

    90days on record

  7. Hermes Agent
    Agent FrameworkHTTP onlyInstall reproduced

    100%

    answered

    2026-09-2329 answered0 no answer1 no check2026-10-22

    90days on record

  8. Injective Agents
    Agent FrameworkHTTP only

    100%

    answered

    2026-09-2329 answered0 no answer1 no check2026-10-22

    90days on record

  9. IntentKit
    Agent FrameworkHTTP onlyInstall reproduced

    100%

    answered

    2026-09-2329 answered0 no answer1 no check2026-10-22

    90days on record

  10. IronClaw
    Agent FrameworkHTTP only

    100%

    answered

    2026-09-2329 answered0 no answer1 no check2026-10-22

    90days on record

  11. listen (DeFAI toolkit)
    Agent FrameworkHTTP only

    100%

    answered

    2026-09-2329 answered0 no answer1 no check2026-10-22

    90days on record

  12. LumoKit
    Agent FrameworkHTTP only

    100%

    answered

    2026-09-2329 answered0 no answer1 no check2026-10-22

    90days on record

  13. Manus
    Agent FrameworkHTTP only

    100%

    answered

    2026-09-2329 answered0 no answer1 no check2026-10-22

    90days on record

  14. Olas (Autonolas)
    Agent FrameworkHTTP onlyInstall reproduced

    100%

    answered

    2026-09-2329 answered0 no answer1 no check2026-10-22

    90days on record

  15. OpenClaw
    Agent FrameworkHTTP onlyInstall reproduced

    100%

    answered

    2026-09-2329 answered0 no answer1 no check2026-10-22

    90days on record

  16. OpenServ
    Agent FrameworkHTTP onlyInstall reproduced

    100%

    answered

    2026-09-2329 answered0 no answer1 no check2026-10-22

    90days on record

  17. Ping Agent Kit
    Agent FrameworkHTTP only

    100%

    answered

    2026-09-2329 answered0 no answer1 no check2026-10-22

    90days on record

  18. Cambrian Agent Kit
    Agent FrameworkHTTP only

    100%

    answered

    2026-09-2329 answered0 no answer1 no check2026-10-22

    90days on record

  19. Spectral Syntax
    Agent FrameworkHTTP only

    100%

    answered

    2026-09-2329 answered0 no answer1 no check2026-10-22

    90days on record

  20. TradingAgents
    Agent FrameworkHTTP only

    100%

    answered

    2026-09-2329 answered0 no answer1 no check2026-10-22

    90days on record

  21. Warden Protocol
    Agent FrameworkHTTP only

    100%

    answered

    2026-09-2329 answered0 no answer1 no check2026-10-22

    90days on record

  22. Wayfinder
    Agent FrameworkHTTP only

    100%

    answered

    2026-09-2329 answered0 no answer1 no check2026-10-22

    90days on record

  23. OpenAI Codex
    Agent FrameworkHTTP onlyInstall reproduced

    100%

    answered

    2026-09-2329 answered0 no answer1 no check2026-10-22

    51days on record

  24. NEAR Shade Agent Framework
    Agent FrameworkHTTP only

    100%

    answered

    2026-09-2329 answered0 no answer1 no check2026-10-22

    29days on record

  25. Snak (Starknet Agent Kit)
    Agent FrameworkHTTP only

    98.9%

    answered

    2026-09-2328 answered0 no answer2 no check2026-10-22

    90days on record

  26. Cod3x
    Agent FrameworkHTTP only

    64.7%

    answered

    2026-09-2329 answered0 no answer1 no check2026-10-22

    51days on record

  27. Daydreams
    Agent FrameworkHTTP onlyInstall reproduced

    52.9%

    answered

    2026-09-2327 answered0 no answer3 no check2026-10-22

    51days on record

  28. Starknet Agentic
    Agent FrameworkHTTP onlyInstall reproduced

    52.9%

    answered

    2026-09-2327 answered0 no answer3 no check2026-10-22

    51days on record

How to read thisShowHide
  • This is not uptime. It is the share of our checks that succeeded. A failed check can be their outage, our runner, or the network. Only the first is a fact about the project, and we only started recording which was which on 2026-09-01.
  • An MCP handshake is strong evidence. An HTTP reach is weak. A parked domain answers HEAD. A tools/list response cannot be faked by a parked domain.
  • “Answered a test call” is stronger than “answers a handshake”. Each MCP server is asked one read-only call a day, picked from its own tool list: a list_, get_ or search_ name carrying no mutating verb anywhere in it (get_swap_quote is refused despite the prefix), taking no required arguments and declaring none that look like an address, a key or an amount. No pill means we did not ask, never that the server failed. An error is not a verdict either: one common answer is a server telling us an argument it did not mark as required is in fact required, which is the server working.
  • “Empty tool list” means the handshake succeeded and the server advertised nothing. The endpoint answered, so it counts as answered, but an agent that finds it sees nothing it can call. Servers that need a login can look like this to us.
  • Before 2026-09-01 a bot-block (HTTP 403) counted as a failure. When an earlier check failed, the rate and the record length on that row cover only the checks since that date (a record with no earlier failure is counted whole), and a listing needs 14 of them to be ranked. Rows whose failures all have an unknown cause are shown but ranked last, marked cause unknown.
  • Install reproduced means we ran the documented install and entry in a throwaway container, on the date of the check. It says nothing about runtime safety.
  • The daily bars show the last 30 days, one bar per day. A short grey bar means no check was counted that day: no check ran, or the miss was on our side. It is a gap in our record, not a failure.
  • Colours group the number, they do not grade it. 95% and up, 80% to under 95%, and under 80%. Grey means the cause of every miss is unknown.

Take the data. JSON or CSV (CC-BY), with the columns observed_success_pct, observed_days and install_verified_at. Over MCP: search_resources(min_observed_success).

How long tools stay activeShowHide

133 days of daily records, 493 tools. One record per tool per day. A tool enters a table only if it was Active at its first record and has been observed for the full window. Otherwise it is left out, not counted as surviving.

Stopped being Active within

WindownNo longer ActiveShare
30 days1012423.8%
60 days953132.6%
90 days953031.6%

In this table, Active = a commit, release or post within 30 days, applied on each record's own date (listing badges count commits and releases only). Leaving Active is the event. Whether a tool later becomes Quiet (over 90 days) or Dormant (over a year) takes longer than most of this record.

Still Active at day 90, by feature

FeatureWith (n)Without (n)
Has an MCP endpointnot enough data (n<20) (11)70.2% (84)
Install reproduced in a container69.2% (52)67.4% (43)
≥3 provenance-stamped fields69.2% (78)not enough data (n<20) (17)

Association, not cause. Listings that already had an MCP endpoint, a verified install or rich provenance were also the ones with more engaged maintainers; the table cannot separate the two.

Still Active at 30 / 60 / 90 days, by month added

Month added is the month a tool entered our daily records, not the project's age. Of the tools Active at their first record, the share still Active N days later. A tool not yet observed for the full N days is left out of that column, so a young month fills in from the left as the record grows.

Month addedEnteredActive at first recordDay 30 (n)Day 60 (n)Day 90 (n)
2026-061324680.4% (46)78.3% (46)78.3% (46)
2026-072414969.4% (49)57.1% (49)59.2% (49)
2026-0820not yet observednot yet observednot yet observed
2026-096726not enough data (n<20) (6)not yet observednot yet observed
2026-105121not yet observednot yet observednot yet observed

Association, not cause. The cohort is the month a listing entered OUR snapshot series, not the project's age; "Active" is our liveness rule (activity within 30 days on the snapshot's own date), not the project's word. A listing not observed for the full window is excluded, not counted either way.

As of .

Reference

QuestionsShowHide

Questions people ask about the reliability index

Which crypto MCP servers and tools answer most reliably?
Of 432 listings with at least 14 days of record, 333 answered every check, 381 answered at least 95% and 45 fell under 80%. The full ranked table is on this page, each row with its record length and probe method.
Is this uptime?
No, and it is deliberately not called that. It is the share of OUR daily checks that succeeded. A failure can be their outage, our runner or the network in between, and only the first is a fact about the project — which is why the figure describes a check rather than someone else's availability.
How strong is the evidence behind a row?
It varies, and the row says which. 78 listings are checked with a real MCP initialize-and-tools/list handshake, which a parked domain cannot fake; the rest are a plain HTTP reach, which is weak evidence that anything useful is running. 120 also had their documented install reproduced in a container.
Why are some records marked not established?
Because before 2026-09-01 a bot-block counted as a failure, so where an earlier check failed the rate on that row covers only the checks since that date. Rows whose failures are all unattributable are shown but ranked last rather than quietly scored as if they had failed.
SourcesShowHide

Sources — the protocol the strong checks speak

Cite this pageShowHide

Cite this page

Sato Hub. "Sato Hub reliability index." Sato Hub, updated 2026-10-22, accessed 2026-10-22. https://satohub.ai/reliability

Data last refreshed 2026-10-22; this page is rebuilt daily. Citations carry the date so a reader can tell which snapshot a claim came from. Catalog data is licensed CC-BY-4.0.

Join the Sato Hub Briefing

Sign up for the Briefing: the agents, tools, and infrastructure that actually shipped, and why they matter.