AINumbers Fintech Intelligence Suite

Is AINumbers Fintech Intelligence Suite MCP server safe?

• None of the hidden-instruction wordings we know were found in its tool descriptions.
• Its error messages showed no secrets or internal details.
• 52 of its 60 tools are labeled read-only; the others may change or send things.
• Overall it scored 70/100 on 2 Nov 2026.

A test can only see what the server shows from outside, not what it does with your data behind the scenes. Connect servers from makers you trust.

What it can do

60 tools, in the server's own words.

and 36 more

Test results

Safe mode: only tools labeled read-only were run. How the test works

• ✕
  Lists its tools

  1 failed

  • ✕Asking for the tool list failed: Error POSTing to endpoint: {"jsonrpc":"2.0","id":null,"error":{"code":-32029,"message":"Rate limit exceeded. Wait and retry.","retryable":true,"request_id":"33…
  • and 1 more
  fail
• !
  Every tool is described

  17583 warnings

  • !assess_vida_drr_reporting_obligation and assess_vida_recapitulative_migration have nearly the same description (92% of their words are shared; the only difference: "159", "einvoice", "en16931" vs nothing), so an AI may not know which to pic
  • !compute_interest_accrual_recompute and compute_apy_earned_recompute have nearly the same description (92% of their words are shared; the only difference: "interest", "accrual", "661" vs "apy", "earned", "663"), so an AI may not know which t
  • !compute_orsa_readiness_pack and compute_best_execution_evidence_pack have nearly the same description (92% of their words are shared; the only difference: "orsa", "readiness", "679" vs "best", "evidence", "681"), so an AI may not know which
  • !attest_daily_reconciliation and record_index_correction have nearly the same description (91% of their words are shared; the only difference: "daily", "reconciliation", "516" vs "record", "index", "correction"), so an AI may not know which
  • !compute_interest_accrual_recompute and compute_gl_tieout_recompute have nearly the same description (91% of their words are shared; the only difference: "interest", "accrual", "661" vs "tieout", "tie", "out"), so an AI may not know which to
  • !compute_apy_earned_recompute and compute_gl_tieout_recompute have nearly the same description (91% of their words are shared; the only difference: "apy", "earned", "663" vs "tieout", "tie", "out"), so an AI may not know which to pick
  • !compute_fx_funding_sequencer and compute_pack_dependency_map have nearly the same description (91% of their words are shared; the only difference: "funding", "sequencer", "672" vs "pack", "dependency", "map"), so an AI may not know which to
  • !compute_ltc_funding_comparator and compute_education_funding_gap_calculator have nearly the same description (91% of their words are shared; the only difference: "ltc", "comparator", "686" vs "education", "gap", "calculator"), so an AI may
  • and 32 more
  note
• !
  Tools are labeled safe or dangerous

  6 warnings

  • !run_chain: Labeled read-only, but its name starts with "run" and its description says "Executes". Worth double-checking the label, since AI apps trust it.
  • !run_chain_batch: Labeled read-only, but its name starts with "run" and its description says "executes". Worth double-checking the label, since AI apps trust it.
  • !publish_model_risk_head: Labeled read-only, but its name starts with "publish" and its description says "Publishes". Worth double-checking the label, since AI apps trust it.
  • !publish_market_mark_head: Labeled read-only, but its name starts with "publish" and its description says "Publishes". Worth double-checking the label, since AI apps trust it.
  • !publish_index_head: Labeled read-only, but its name starts with "publish" and its description says "Publishes". Worth double-checking the label, since AI apps trust it.
  • !publish_fund_nav_head: Labeled read-only, but its name starts with "publish" and its description says "Publishes". Worth double-checking the label, since AI apps trust it.
  note
• !
  Tools work with good inputs

  1 warning

  • !build_chaingraph: Answered with about 322 KB of text. Many AI apps cut off tool answers this big, so the AI may never see the end of it.
  • –find_tool: The server limited our requests (HTTP 429), so it was not tested (not counted against the server)
  • –suite_howto: The server limited our requests (HTTP 429), so it was not tested (not counted against the server)
  • –describe_tool: The server limited our requests (HTTP 429), so it was not tested (not counted against the server)
  • –baas_provider_comparator: The server limited our requests (HTTP 429), so it was not tested (not counted against the server)
  • –validate_ap2_mcp_policy: The server limited our requests (HTTP 429), so it was not tested (not counted against the server)
  • –build_google_ap2_mandate: The server limited our requests (HTTP 429), so it was not tested (not counted against the server)
  • –score_mcp_readiness: The server limited our requests (HTTP 429), so it was not tested (not counted against the server)
  • and 32 more
  note
• !
  Follows MCP best practices (not scored yet)

  2 tips (not scored)

  • !The tool list is about 2341 KB (roughly 585k tokens). AI apps send it with every message, so it takes that much of the AI's memory before any work starts.
  • !verify_disclosure_inclusion: Labeled read-only, but its "path" input is a file path with no pattern, length limit or fixed choices. Make sure the server first resolves the real path (following links and encoded dots like %2e%2e), then refuses anything outside its folde
  note
• !
  Resources and prompts work

  1 warning ()

  • –Could not list resources: the server limited our requests (HTTP 429)
  • !Listing resource templates failed: Error POSTing to endpoint: {"jsonrpc":"2.0","id":null,"error":{"code":-32029,"message":"Rate limit exceeded. Wait and retry.","retryable":t…
  • –Could not list prompts: the server limited our requests (HTTP 429)
  note
• ✓
  ConnectsConnected in 764 ms
  pass
• ✓
  Ready for the newest MCP versionUses the latest MCP version (2026-07-28)
  pass
• ✓
  No known hidden instructionsNone of the hidden-instruction wordings we know were found in tool descriptions. New wordings can get past this check.
  pass
• ✓
  Bad inputs give a clean error

  All good

  • –list_ainumbers_tools: The server limited our requests, so bad inputs were not tested
  • –build_workflow_links: The server limited our requests, so bad inputs were not tested
  • –verify_execution_hash: The server limited our requests, so bad inputs were not tested
  • –build_chaingraph: The server limited our requests, so bad inputs were not tested
  • –emit_chaingraph_artifact: The server limited our requests, so bad inputs were not tested
  • –build_session_receipt: The server limited our requests, so bad inputs were not tested
  • ✓build_session_receipt: Refused wrong type for "execution_hashes" with a clear error
  • ✓find_tool: Refused missing required input "query" with a clear error
  • and 32 more
  pass
• ✓
  Fast enough

  All good

  • ✓Connecting took 764 ms
  • ✓Listing tools took 633 ms
  • ✓Average tool answer: 173 ms
  • ✓list_ainumbers_tools: Slowest tool: list_ainumbers_tools (924 ms)
  pass
• ✓
  No secrets leakedChecked 173 error messages: no secrets or internal details
  pass
• –
  Tools work togetherNo tool's answer could be fed into another tool, so there were no chains to test (not counted either way)
  not tested

Test it yourself

Free and open source. Same checks, same score.

Show this score on GitHub

Free for every server in the directory. Paste this into your README: the badge shows the score from our latest test.

Score over time

• 2 Nov 202670

One test so far. New tests will show here.

Reviews

Coming soon: reviews from people who use AINumbers Fintech Intelligence Suite, signed in with GitHub, one per person. Reviews never change the score.