MCP-сервер ScrubMyText

• Имя в реестреcom.scrubmytext/toolsподтверждён в реестре MCP

Имя в реестре лежит в пространстве scrubmytext.com. Реестр пускает в такое пространство только того, кто доказал владение доменом через DNS или HTTP.

• Состояниеотвечаетпроверено 2026-10-25
• Инструменты49
• Сервер называет себяcom.scrubmytext/tools
• Версия сервера7.5.0
• Версия протокола2025-11-25

2026-10-25 сервер ответил на initialize и tools/list без входа. Первый ответ за 252 мс.

• Инструменты делают все проверенные сценарии: scan documents for likely secrets before a model reads them; check text for hidden instructions before an agent acts on it; stop an agent repeating a payment after a timeout retry.
• Проверок пока 3 из 7, нужных, чтобы судить о надёжности.

Вывод сделан правилами по замерам выше. Вероятный первый вызов выбрала модель по списку инструментов: прочитайте описание, прежде чем на него полагаться.

• Оценка INITE MCPчерез неделюдля оценки нужно 7 ежедневных проверок, пока 3
• Что агент может с ним сделать3 из 3 сценариев
• Отвечалидёт замерпроверок пока 3, долю покажем после 7
• Медиана первого ответа97 мс
• Место в контексте агента≈ 8 082 токеновописания всех инструментов (49), они занимают контекст агента на каждом ходу
• Описания инструментовB · 83 из 100
• Типизированные ответы2 из 49 инструментовобъявлен outputSchema: ответ приходит с типами и читается сразу

• Параметры без типа: в 1 из 49
• Параметры без описания: в 48 из 49

Программа проверяет сервер раз в день: initialize и tools/list, без модели. Токены - это длина описаний из расчёта примерно четыре символа на токен. Оценка - среднее шести проверок имени, описания, схемы и аннотаций каждого инструмента. Она говорит, насколько понятно описаны инструменты, а не о том, работают ли они. Оценка складывается только из замеренного: отвечает ли сервер (25%), делает ли работу продукта (25%), описания (20%), контекст (10%), использование (10%) и насколько просто подключиться (10%). Незамеренная часть не учитывается, а скрытая инструкция ограничивает оценку 40 баллами. Состояние MCP-серверов вендоров: методика и цифры →

• scrub_text

  Normalize hidden/control characters and formatting artifacts without rewriting wording. Free or paid.

• inspect_text

  Inspect exact Unicode, invisible characters, normalization, and mixed-script anomalies. Free or paid.

• scan_untrusted_text

  Use before an agent ingests pasted, retrieved, uploaded, or externally supplied text. Returns structural and heuristic risk signals; it does not determine maliciousness or safety. Trial, subscription, or x402 pay-per-call.

• redact_sensitive_text

  Use before storing, sharing, logging, or sending text downstream when it may contain credentials or identifiers. Pattern-based redaction; trial, subscription, or x402 pay-per-call.

• compare_text

  Use to verify what changed between two text versions and whether differences are only scrub/format artifacts. Trial, subscription, or x402 pay-per-call.

• detect_secrets

  Use before an agent logs, stores, or forwards text that may contain credentials. Finds likely API keys, private keys, JWTs, cloud credentials, and secret patterns without echoing values. Trial, subscription, or x402 pay-per-call.

• validate_output

  Use immediately before a downstream action when model output must be valid JSON or match a schema. Trial, subscription, or x402 pay-per-call.

• sanitize_agent_context

  Use before putting untrusted retrieved or uploaded text into a model context. Normalizes structural artifacts, returns risk signals, and can redact sensitive patterns. Trial, subscription, or x402 pay-per-call.

• prepare_for_agent

  Use as the default preprocessing step before an AI agent consumes untrusted external text. Returns agent-ready text plus inspection, risk-scan, and secret-detection reports. Trial, subscription, or x402 pay-per-call.

• rework_text

  Use when an agent needs a deterministic, no-LLM rewrite while preserving protected facts such as numbers, URLs, emails, and quotes. Returns change metrics and preservation checks; does not verify authorship or watermark status. Trial…

• fingerprint_action

  Use before retryable work to derive a stable SHA-256 action fingerprint and idempotency key from an action name, optional scope, and JSON arguments. Object key order does not affect the result. Returns no input values and stores nothing…

• lock_action

  Use immediately before a consequential agent action to atomically claim an idempotency key so retries cannot execute it twice. Subscription API key required.

• lock_action_intent

  Use immediately before a consequential retryable action when no stable idempotency key exists. Deterministically fingerprints the action and arguments, atomically claims that key, and returns safe_to_execute. Repeat calls with equivalent…

• check_action

  Use to inspect an idempotency key before or after an agent action. Subscription API key required.

• complete_action

  Paid: mark the current lock completed and preserve the idempotency record for its configured retention period.

• release_action

  Paid: release an uncompleted action lock using its lock_id so another attempt can safely acquire it.

• create_signal

  Use when an agent must wait for an external callback, webhook, job completion, or asynchronous event. Subscription API key required.

• get_signal_events

  Paid: retrieve events received by a temporary signal inbox in arrival order.

• delete_signal

  Paid: immediately delete a signal inbox and all stored event bodies.

• request_approval

  Use when an agent needs independent human approval before a consequential action. Creates a secure approval link. Subscription API key required.

• check_approval

  Paid: check whether a human approval request is pending, approved, rejected, expired, or cancelled.

• cancel_approval

  Paid: cancel a pending human approval request so its link can no longer approve the action.

• manage_action_challenge

  Create, inspect, consume, or cancel a digest-bound cooling-off window. A bearer challenger can veto before the deadline; afterward the owner can atomically consume one execution right during a bounded window. This is exception handling…

• reserve_action_budget

  Atomically reserve integer units under a named, period-scoped ceiling. A reservation_key makes retries idempotent. The service stores hashes and counters, not purchase details. Production; subscription API key required.

• record_action_receipt

  Create or retrieve an immutable receipt digest for an action outcome without storing the action or outcome payload. Production; subscription API key required.

• verify_signed_signal

  Verify an HMAC-SHA256 signal, enforce timestamp freshness, and reject a replayed nonce. The message and shared secret are never stored. Production; subscription API key required.

• manage_resource_lease

  Atomically acquire, inspect, or release a short-lived exclusive lease on a hashed resource key. Production; subscription API key required.

• manage_authority_delegation

  Issue, consume, inspect, or revoke a narrow bearer capability bound to one action fingerprint and a maximum use count. This is a technical capability, not identity proof or a legal delegation. Production; subscription API key required.

• manage_agent_stop

  Set, inspect, or clear a durable stop signal for a hashed agent or workflow key. Callers should fail closed whenever stopped is true. Production; subscription API key required.

• manage_task_handoff

  Create a bearer-token task handoff, let another paid API identity claim it exactly once, and inspect, complete, or cancel the transfer. Only task fingerprints are stored. Production.

• manage_action_reconciliation

  Open, inspect, append digest-only evidence to, or resolve a discrepancy between expected and observed action outcomes. Production; subscription API key required.

• anchor_context_trace

  Create or inspect an immutable provenance anchor from a context digest, parent digests, and processing stage. No context content is stored. Production; subscription API key required.

• manage_task_contract

  Offer, accept, inspect, complete, or cancel a digest-bound technical task commitment. This records service state; it is not a legal contract, escrow, payment, or identity verification. Production.

• check_agent_readiness

  Deterministically check a supplied MCP server card against a narrow, published integration contract. The service does not fetch arbitrary URLs and does not certify security or correctness. Free or paid.

• manage_action_quorum

  Create, inspect, consume, or cancel an N-of-M bearer authorization quorum. Each unique bearer can decide once; only a successful one-time consume grants an execution right. Subscription API key required.

• manage_task_deadline

  Create, inspect, complete, or cancel a task deadline whose due and escalation state is derived from an independent server clock even when the agent is offline. Subscription API key required.

• manage_attention_budget

  Create, inspect, reserve, release, or close a finite review-capacity pool with per-subject caps, retry-safe reservations, and protected priority capacity. Subscription API key required.

• manage_agent_escalation

  Create, inspect, or cancel a digest-only incident with a private bearer operator link and deterministic response and resolution SLA state. Link delivery remains the caller’s responsibility. Subscription API key required.

• manage_terms_pin

  Pin, inspect, or revoke an immutable digest-bound record of observed terms, interface, version, and price. Stores bounded metadata and digests, not the terms body. This is technical evidence, not a legal agreement. Subscription API key…

• match_policy_requirements

  Deterministically compare bounded provider capabilities with consumer-required and forbidden policy tokens. Processes supplied tokens without storing policy content and does not make legal or compliance conclusions. Subscription API key…

• manage_action_compensation

  Bind a compensating-action digest before execution, record that the primary action executed, and atomically mark compensation once. Stores digests only and does not perform either action. Subscription API key required.

• manage_context_status

  Publish, inspect, supersede, dispute, or revoke digest-only context status so downstream agents can stop using stale inputs. Stores no context body and does not invalidate copies outside this service. Subscription API key required.

• manage_context_rights

  Issue, inspect, evaluate, or revoke a digest-bound machine-readable context-rights envelope. It is a technical policy signal, not legal advice or proof of ownership. Subscription API key required.

• manage_disclosure_budget

  Create, inspect, record against, or close a cumulative disclosure budget with total, per-recipient, and per-category limits. Stores hashes and counters, not disclosed content or recipient identities. Subscription API key required.

• list_quality_cards

  List exact service-version/task quality cards that have reached the public three-verified-contributor evidence threshold. Results are ordered by recent evidence, not quality, payment, or sponsorship. Free.

• get_quality_card

  Retrieve a public, contextual quality card for an exact MCP server or REST API version. Scores remain unavailable until at least three independent verified contributors submit structured observations. Free.

• compare_quality_targets

  Return contextual quality cards for two to ten exact targets in caller-provided order. It exposes evidence and uncertainty without producing a paid ranking or universal winner. Free.

• record_quality_observation

  Submit an opt-in structured observation for an exact target and task. Receipt-linked observations are labeled E1; other observations are E0. Stores no prompt, response, credential, email, or free-form review. Free verified contributor or…

• request_quality_correction

  Attach a digest-only correction or appeal request to a quality card. Requests are visible as unverified pending state and do not erase evidence or prove provider identity. Free verified contributor or subscription key required.