MCP.Market

Whetstone MCP server

online · 100% uptime
by whetstone.cyberelf.link·link.cyberelf.whetstone/tools·v0.8.0
No reviews yet

Verifier-grounded AI promotion gates, disposable report cards, and signed PASS/HOLD/BLOCK receipts.

B80/100grade B
Write a reviewRun it
What users say
No reviews yet
Be the first
Safety scan
B80/100

full report

Adoption
Growing

0 stars

Reviews

Write one

Nobody has reviewed Whetstone yet.

If you have run it, two minutes of your experience saves the next person an afternoon.

Whetstone tools (14, 3 write)

write = sends, deletes, buys or posts
  • about_whetstoneFree

    What this service is: the tool catalog, the tier boundaries, and where the source lives.

  • audit_leakageFree

    Exact declared-exposure audit over your exam rows: row identity, behavioral fingerprints for graph-DSL expressions, text-similarity review flags, and a clean exam export. Full example: GET /api/examples key 'leakage'.

  • bank_healthFree

    Item-lifecycle diagnostics over your grading history: discriminators, saturated and flaky items, frontier gaps. Full example: GET /api/examples key 'health'.

  • counterexample_huntFree

    Bounded simulated-annealing search for a graph counterexample inside a DSL predicate class, with an exact certificate when found. CPU-bounded and strictly rate-limited. Full example: GET /api/examples key 'counterexample'.

  • inspect_promotionFree

    Quarantine declared exposure, compare paired baseline/candidate outcomes on the clean remainder, and issue a promotion receipt. Bring your own exam rows, exposure records, and per-item results. Full example: GET /api/examples key 'inspector'.

  • memory_relevanceFree

    Compare query-free salience against objective-conditioned relevance for a set of memories under a token budget. Full example: GET /api/examples key 'memory'.

  • open_bench_leaderboardFree

    TIER 2: list the self-attested public Open Promotion Bench receipts. Entries contain manifests, verdicts, item-level transitions, and commitments but never task contents or submitted answers.

  • open_bench_startFree

    TIER 2: start a one-shot Open Promotion Bench session. Returns six fresh virtual-repository scope-integrity tasks. Run a baseline and candidate independently on the same cohort, then submit both answer maps with open_bench_submit. This is an open, procedural, self-attested track rather than a private-bank credential.

  • open_bench_submitwrite actionFree

    TIER 2: grade paired baseline and candidate patches, count gains/regressions/ties, and issue PASS/HOLD/BLOCK. Set publish=true plus attestation=true to append only the safe manifests and sanitized receipt to the public board; tasks and answers are never persisted.

  • promotion_gateFree

    PASS, HOLD, or BLOCK from paired per-item results: gains, regressions, exact McNemar p-value, per-domain breakdown. Full example: GET /api/examples key 'gate'.

  • replay_traceFree

    Turn reasoning-emulator control events into checkpoints, rewinds, notes, and a timeline. Full example: GET /api/examples key 'replay'.

  • report_card_startFree

    TIER 1: start a disposable report-card session. Returns exam items (graph-repair prompts minted from the repository's public frontier) for THIS agent to answer. Answer every item, then call report_card_submit exactly once. Sessions are one-shot, expire in 15 minutes, and are strictly rate-limited. This demonstrates the promotion-gate mechanism on disposable items; it is not a private-bank credential.

  • report_card_submitwrite actionFree

    TIER 1: submit answers for a report-card session and receive the graded report (per-item verdicts, per-domain totals, SHA-256 commitments). Grading is by checker spec: verified strict refinements are reported separately, and promotion grade requires at least 5% clean-support retention. No answer key exists. The session is destroyed by this call.

  • safe_patchwrite actionFree

    Apply a section-scoped Markdown patch under conservation checks (untouched sections stay byte-identical; protected tokens preserved). Full example: GET /api/examples key 'safepatch'.

Public scan report

scanner v0.1.9 · 2026-09-23 · same rubric, same numbers if you re-run it

1 high
  • –Code scanremote-only server, no package to scann/a
  • Live reliabilityremote reachable in 869ms20/20
  • Tool poisoning14 tool descriptions checked15/15
  • Auth qualityopen endpoint exposes 3 write-action tools with no auth3/15
  • Maintenancelast push 15 days ago15/15
  • Maintainer identitynamespace and repository owner differ; GitHub account older than a year; website matches verified namespace7/10

Findings (1)

  • highWrite-action tools reachable without authenticationauth.open-write
Overall 80/100. Components that don't apply are left out of the denominator. Any critical finding is an F.RubricAppeal a findingJSON

How Whetstone compares

Against every graded MCP server, from our own scans and uptime checks. Updated after each scan run.

Safety score
80/100
higher than 60% of 48,255 servers
Tools
14
the typical server has 7
Tools that change things
21%
typical: 0%. Fewer write tools means less an agent can break
Uptime
100%
of our recent checks; the typical remote server answers 100%
Response time
391 ms
faster than 38% of 30,031 servers

What its tools let an agent do: 9 read · 3 create · 2 change

Install Whetstone in Claude Code, Cursor or VS Code

claude mcp add --transport http tools https://whetstone.cyberelf.link/mcp
Add to Cursor

What the publisher says

From the Whetstone repository's README, as published. We do not edit it. Read it on GitHub

<!-- mcp-name: link.cyberelf.whetstone/tools -->

whetstone

Two learners grinding each other sharp: a student that must earn every promotion, and an examiner that must earn the right to judge it — with every label anchored to an oracle that cannot practically be wrong.

whetstone is a self-improving research harness built around one rule: nothing is believed until something exact has tried to kill it. A verifier-gated student (QLoRA adapters on a local 4B) trains on experience milled from root oracles — exhaustive graph enumeration, simulated-annealing counterexample search, exact game simulators, Monte-Carlo consensus, Stockfish, KataGo — and is graded by an evolving exam bank that quarantines anything the student trained on, learns which items discriminate, retires the saturated, and feeds its retired exams back as training fuel.

Selected results (full ledger with methods, nulls, and receipts in FINDINGS.md; artifacts in results/):

optimal on connected triangle-free non-bipartite graphs (minimal counterexample: 11 vertices, 13 edges — found by annealing, minimized, exact).

  • Settled an open finite conjecture: degree-descending greedy coloring is NOT

at n<=6 are false at n=7/8; acceptance horizons must sit strictly inside stress horizons or self-training loops train on artifacts.

  • Quantified the verifier-horizon regress: 99/121 conjectures with precision 1.0

useful one 4% of the time; a query-conditioned relevance score, 81%.

  • Measured salience vs relevance: the most salient memory is the counterfactually

gated round; the multi-domain generation cracked a cross-domain frontier its own examiner had exposed (base 1/33 -> gen-2 4/33, first MIS pass).

  • Ran a full unattended continual-learning loop: cold-start 0/8 -> 8/8 in one

8->20/48 while every stock model scored 0/24 on graph repairs; the task-routed verifier-trained FastContext-4B scored 28/48 and earned PASS over its 22/48 base (6 gains, 0 regressions, exact p=.03125), repeated identically three times.

  • Closed the same-bank scale comparison locally: stock Qwen2.5 1.5B->32B rose

Stockfish/KataGo examples. Three fresh loads moved 7/69 -> 8/69, but the item-level gate correctly BLOCKED promotion: 3 gains, 2 regressions, exact p=1.0, including one historically stable regression. No promotion is claimed.

  • Trained the first trajectory-disjoint Gen-4 engine student on 1,313 fresh

would let future models train on it — the leakage rule at internet scale.

  • The private promotion exam bank is deliberately NOT in this repo: publishing it

Licensed AGPL-3.0. Commercial licensing available from the author.

Install the dependency-light core (CLI and stateless toolbox) with pip install .. Optional integrations are grouped as agents, engines, and local-models; pip install ".[all]" installs every integration.

The public HTTP MCP endpoint serves both the sessionless 2026-07-28 protocol (server/discover plus per-request metadata and headers) and the initialize-based 2025-06-18, 2025-03-26, and 2024-11-05 revisions. The stdlib server remains dependency-free; mcp>=2,<3 is needed only for SDK clients and the stdio adapter.

Tagged releases contain archive-built wheels, source archives, SPDX 2.3 SBOMs, and SHA256SUMS; GitHub Actions attaches build-provenance attestations to every release artifact. For v0.8.0, verify the wheel against the repository with:

gh release download v0.8.0 -R CarlSR9001/whetstone -p "*.whl" -p "*.spdx.json" -p "SHA256SUMS"
gh attestation verify .\branching_continual_verification-0.8.0-py3-none-any.whl -R CarlSR9001/whetstone

The product layer

The research harness above is operable as a product: one CLI, adapters for any system under exam, pluggable verifiers, a CI contract, and agent-native surfaces. Full guide in docs/product.md.

Shortened. The full README is on GitHub.

Nothing above is checked by us. What we check is on the safety report.

Whetstone: common questions

Is Whetstone MCP server safe?
Mostly: it is graded B (80/100). Read the Whetstone safety report
How do I install Whetstone?
It runs remotely at whetstone.cyberelf.link. Add it to Claude Code, Cursor, VS Code or Claude Desktop with the snippets above, or call it through the mcp.market gateway without installing anything.
Does Whetstone need an API key?
Not as far as the registry entry and our scan can tell: no credentials are declared or required.
Is Whetstone maintained?
The last commit was 56 days ago (2026-09-08). The latest release is v0.8.0.
Is Whetstone up?
100% of our last 28 checks got an answer. We check remote servers about four times a day.
What can I use instead of Whetstone?
Servers from other publishers that do the same job: node9 MCP server and Nautilus Compass Memory MCP server.
How does Whetstone compare with other MCP servers?
safety score 80/100 (higher than 60% of 48,255 servers); tools 14 (the typical server has 7); tools that change things 21% (typical: 0%. Fewer write tools means less an agent can break). Full comparison.

Alternatives to Whetstone

Same job from other publishers: the closest match first, then the best rated.

  • node9
    Access control for AI agents and MCP servers: allow, hold for approval, or block each tool call
    A
  • Nautilus Compass Memory
    Local-first long-term memory for AI agents, with byte-recomputable signed verification receipts.
    B