Сервер MCP, который позволяет любому ИИ-агенту проверять и улучшать простоту языка своего текста — до того, как этот текст попадёт к человеку. Полностью локально: не нужен API-ключ, не требуется сетевой вызов, нет внешнего сервиса. Это чисто вычислительный инструмент, поэтому всё, что вы через него пропускаете, никогда не покидает машину, на которой он запущен.
В отличие от других инструментов этой серии (scamlens, clearread, clauselens), это не сайт, который открывает человек, — это инфраструктура для самой экосистемы ИИ-агентов. Любой MCP-совместимый клиент (Claude Code, Claude Desktop или любой другой) может вызывать его напрямую.
Агенты генерируют много текста — ответы, документацию, уведомления, сообщения об ошибках — и в большинстве случаев не имеют возможности проверить, действительно ли этот текст легко читается получателем, прежде чем отправить его. Этот сервер даёт каждому агенту такую самопроверку как полноценный вызов инструмента, используя тот же механизм простого языка (разбиение предложения + выверенный словарь замены сложных слов на простые), который уже включён и проверен в clearread — автономном веб-приложении из этой серии. Здесь он доступен для программного использования вместо ручной вставки в браузер.
Related MCP server: readability-claude
readability_score_textВычисляет Flesch Reading Ease и Flesch open Flesch-Kincaid Grade Level для блока английского текста. Для нелатинской письменности возвращает supported: false и только структурные счётчики, поскольку формулы Flesch предполагают английские слоговые модели.
readability_simplify_textПереписывает текст более простым языком: разбивает слишком длинные предложения на естественные границы придаточных и заменяет распространённую бюрократическую лексику на более простые эквиваленты. Сообщает оценку Flesch Reading Ease до и после, чтобы вы могли убедиться, что переписывание действительно помогло. Только английский текст — это осознанное проектное решение: подстановка слов ненадёжна в языках с богатой морфологией, поэтому нелатинский текст обнаруживается и передаётся без изменений, чтобы не рисковать поломкой грамматики. См. также после примечание в src/services/simplify.ts, почему несколько заманчивых словарных статей, например «accompanied» → «went with», были исключены после того, как они нарушили устойчивые фразы типа «must be accompanied by».
Склонируйте и соберите:
git clone https://github.com/wedo911/readability-mcp-server.git
cd readability-mcp-server
npm install
npm run buildДобавьте его в конфигурацию вашего MCP-клиента (например, claude_desktop_config.json или под проектируемого . mcp.json для Claude Code):
{
"mcpServers": {
"readability": {
"command": "node",
"args": ["/absolute/path/to/readability-mcp-server/dist/index.js"]
}
}
}npm run build
node --test tests/textStats.test.mjs tests/simplify.test.mjsMCP Inspector может вызывать инструменты прямо из командной строки:
npx @modelcontextprotocol/inspector --cli node dist/index.js \
--method tools/call --tool-name readability_simplify_text \
--tool-arg text="Please utilize the enclosed form prior to the deadline."MIT — см. LICENSE.
Compute Flesch Reading Ease and Flesch-Kincaid Grade Level for a block of English text.
Use this to check how easy your own draft output (a reply, a document, documentation) will be for a reader to understand before sending it -- useful for agents that want to self-check plain-language quality, or for comparing readability across different drafts.
Args:
text (string, 1-20000 chars): The text to score.
response_format ('markdown' | 'json'): Output format (default: 'markdown').
Returns: For JSON format: { "supported": boolean, // false if the text contains non-Latin script (Flesch formulas assume English) "fleschReadingEase": number|null, // 0-100, higher = easier; null if unsupported or no words found "fleschKincaidGradeLevel": number|null, // approximate US school grade level "gradeLevelLabel": string|null, // human description, e.g. "standard (around 8th-9th grade)" "sentenceCount": number, "wordCount": number, "syllableCount": number, "avgWordsPerSentence": number, "avgSyllablesPerWord": number }
Examples:
Use when: "Is this support-ticket reply easy enough to understand?" -> score the reply text
Use when: "Which of these two drafts reads more simply?" -> score both, compare fleschReadingEase
Don't use when: the text is not primarily English -- scoring is skipped for non-Latin script and only structural counts are returned
Error Handling:
Returns an error if text is empty or exceeds 20000 characters (split long documents into sections first).
| Name | Required | Description | Default |
|---|---|---|---|
| text | Yes | The text to score, e.g. a draft reply, an email, or documentation. | |
| response_format | No | Output format: 'markdown' for a human-readable summary or 'json' for structured data. | markdown |
Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?
The description discloses important behavior beyond the read-only/idempotent annotations: non-Latin text yields supported=false, metric values can be null, only structural counts are returned for unsupported input, and empty or over-long text raises an error. It also documents the JSON return shape in detail, which matters because no output schema is present.
Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.
Is the description appropriately sized, front-loaded, and free of redundancy?
The description is front-loaded with a precise purpose and then organized into useful sections: usage, args, returns, examples, and errors. The Args section duplicates schema details, but the rest of the content earns its place because it covers usage scenarios, fallback behavior, and failure conditions.
Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.
Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?
For a small, low-risk tool with no output schema, the description is unusually complete: it covers supported inputs, output metrics with units and null semantics, error conditions, and common use cases. There is no crucial operational gap for an agent deciding whether to call and how to interpret results.
Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.
Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?
Schema coverage is 100%, so the schema already documents both parameters and their defaults. The description mostly restates the parameter constraints and adds useful-but-not-critical context such as splitting long documents; it does not meaningfully increase parameter-level understanding beyond the schema.
Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.
Does the description clearly state what the tool does and how it differs from similar tools?
The opening sentence names a specific operation—compute Flesch Reading Ease and Flesch-Kincaid Grade Level for a block of English text—rather than restating the tool title. The motivating examples ('check a draft reply', 'which draft reads more simply') make its job and distinction from a simplification tool clear.
Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.
Does the description explain when to use this tool, when not to, or what alternatives exist?
The description gives concrete 'use when' scenarios and an explicit 'don't use when' condition for non-English text, which is strong guidance for tool selection. It does not explicitly point to readability_simplify_text as the alternative, but the separation of scoring vs. simplifying is largely implied.
Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.
Rewrite English text in plainer language: breaks overly long sentences at natural clause boundaries and swaps common bureaucratic vocabulary ("utilize" -> "use", "prior to" -> "before") for plainer equivalents. Also reports the Flesch Reading Ease score before and after, so you can confirm the rewrite actually helped.
Use this before sending a reply, document, or notice to a user who may benefit from plainer language -- including as a self-check step for an agent's own drafted output.
Args:
text (string, 1-20000 chars): The English text to simplify.
response_format ('markdown' | 'json'): Output format (default: 'markdown').
Returns: For JSON format: { "supported": boolean, // false if text contains non-Latin script (English-only feature); simplifiedText equals the input unchanged in that case "simplifiedText": string, "changeCount": number, // total words/phrases substituted "changes": [ { "start": number, "end": number, "original": string, "simplified": string } ], // offsets index into simplifiedText "truncated": boolean, // true if changeCount exceeds 50 (only the first 50 are listed) "fleschReadingEaseBefore": number|null, "fleschReadingEaseAfter": number|null }
Examples:
Use when: "Make this notice easier to read" -> pass the notice text
Use when: checking whether your own drafted reply is needlessly complex before sending it
Don't use when: the text is not primarily English -- word substitution is skipped for non-Latin script (Arabic morphology in particular makes naive word-swapping unreliable) and the text is returned unchanged, with supported=false
Error Handling:
Returns an error if text is empty or exceeds 20000 characters.
| Name | Required | Description | Default |
|---|---|---|---|
| text | Yes | The English text to simplify, e.g. a draft reply before sending it to a user. | |
| response_format | No | Output format: 'markdown' for a human-readable summary or 'json' for structured data. | markdown |
Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?
Beyond the annotations declaring it read-only and idempotent, the description discloses important behaviors: non-Latin scripts return supported=false with unchanged text, only the first 50 changes are listed when changeCount exceeds 50, and Flesch scores are reported before and after. It also explains error conditions for empty or overly long input.
Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.
Is the description appropriately sized, front-loaded, and free of redundancy?
The description is well-structured and front-loaded: the core behavior appears in the first sentence, followed by usage guidance, argument details, return structure, and error handling. Each section serves a clear purpose and the length is justified by the richness of the behavior being described.
Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.
Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?
With no output schema present, the description compensates by listing the full JSON return shape, including edge-case fields like truncated and supported. It also covers input constraints, non-English behavior, and concrete usage examples, leaving an agent well-equipped to format inputs and interpret outputs correctly.
Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.
Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?
The schema already covers both parameters 100%, but the description adds value by documenting the JSON response structure, field semantics, and the default response_format. It also gives context for the text parameter through examples like passing a draft reply.
Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.
Does the description clearly state what the tool does and how it differs from similar tools?
The description states a specific action and resource: rewriting English text in plainer language, breaking long sentences, and substituting bureaucratic vocabulary. It also mentions the Flesch Reading Ease output, which helps distinguish it from the sibling scoring tool by emphasizing that the core job is simplification, not mere scoring.
Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.
Does the description explain when to use this tool, when not to, or what alternatives exist?
The description gives explicit use cases, such as simplifying a response, document, or notice, and also works as a self-check for the agent's own drafted output. It gives a clear when-not-to-use case for non-Latin script, though it does not explicitly name the sibling readability_score_text as the alternative when only a score is needed.
Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
v1.0.0readability_score_textreadability_simplify_textScored across 2 tools
The two tools have clearly distinct purposes: one measures readability metrics, the other rewrites text to simplify it. There is no overlap or ambiguity in choosing between them.
Both tool names follow a consistent pattern: domain prefix 'readability_' plus verb_noun pairs ('score_text', 'simplify_text'). The naming is uniform and predictable.
Two tools is on the thin side for a server, but it is a tightly focused scope: measuring and simplifying readability. It feels slightly minimal but each tool earns its place.
For the stated purpose of assessing and improving plain-language readability, the server covers both key operations: evaluating a text's readability and rewriting it into plainer language. No obvious critical gap exists.
Clean Markdown and AI-readability scoring for any URL. Built for AI agents.
The open, measured register where AI agents evolve written English together.
Check text before your agent reads it: hidden instructions, invisible text and links that leak data.
Toxicity scores (0-1) for text before an agent posts it. Perspective API compatible. No LLM.
