A
licenseA
qualityD
maintenanceEnables AI attribution audits and quick transparency checks on written content, providing authorship analysis and segment breakdowns for editorial review.
2
2
MIT
avoid-ai-writing-mcp
This server provides a local, deterministic, read-only way to audit prose for AI-writing patterns without sending text anywhere or using a language model.
score_text: Returns a compact 0–100 heuristic pattern score, classification (UNSCORED/HUMAN_ONLY/MIXED/AI_ONLY), confidence, probabilities, word count, issue count, and context.audit_text: Returns the same score plus up to 100 flagged pattern issues with suggested alternatives, up to 100 highlighted sentence regions, statistics, and truncation counts.Choose an optional context:
general(default),technical,marketing, orpersonalto tune detection, withtechnicalreducing noise from code-adjacent prose.Handles text up to 100,000 characters, requires at least 1 character, and reports reasons when text cannot be scored (empty, too short, or too long).
Runs entirely over local stdio; no network calls, no API, no open ports, and no rewriting tools.
Click on "Deploy Server".
Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state.
In the chat, type
@followed by the MCP server name and your instructions, e.g., "@avoid-ai-writing-mcpCan you check this text for AI writing patterns?"
That's it! The server will respond to your query, and you can continue using it as needed.
Here is a step-by-step guide with screenshots.
avoid-ai-writing-mcp
A local, deterministic MCP server for auditing prose with the published
avoid-ai-writing-detector.
It exposes two read-only tools:
score_textreturns a compact score, classification, confidence, and counts.audit_textadds a bounded set of flagged patterns, suggested alternatives, statistics, and highlighted sentence regions.
The server makes no network calls and uses no language model. Text stays inside the local MCP process. Scores are heuristic writing-pattern signals, not proof that a person or model wrote the text.
Requirements
Node.js 20 or newer
An MCP host that supports local stdio servers
Related MCP server: Byakugan: Private, open-source AI-text checker
Install
Add this server to Claude Code:
claude mcp add avoid-ai-writing -- npx -y avoid-ai-writing-mcpOr add it to an MCP JSON configuration:
{
"mcpServers": {
"avoid-ai-writing": {
"command": "npx",
"args": ["-y", "avoid-ai-writing-mcp"]
}
}
}To run the current source directly from GitHub instead, use
github:conorbronsdon/avoid-ai-writing-mcp as the package spec.
The package uses stdio for protocol messages. It does not open a port or send input to an API.
Tool inputs
Both tools accept:
Field | Type | Required | Description |
| string | yes | Text to evaluate locally (maximum 100,000 characters). |
| enum | no |
|
technical mode reduces noise from patterns that are normal in code-adjacent
writing. The other modes preserve the detector's context-specific behavior.
audit_text returns at most 100 issues and 100 highlighted regions. Its
truncated field reports how many additional items the detector produced.
Development
npm install
npm test
npm run inspectnpm run inspect launches the official MCP Inspector against the local stdio
entry point.
Scope
This package intentionally does not expose a rewrite tool. Rewriting requires editorial judgment and often a hosted model; adding that would undermine this server's deterministic, local, privacy-preserving contract.
License
MIT
Available Tools
2 toolsaudit_textAudit text for AI-writing patternsARead-onlyIdempotent
Audit text locally with the deterministic Avoid AI Writing detector. Returns the score plus up to 100 flagged patterns and 100 highlighted sentence regions, with truncation counts. This is a heuristic writing audit, not proof of authorship; use score_text for a compact result.
ParametersJSON Schema
| Name | Required | Description | Default |
|---|---|---|---|
| text | Yes | Text to evaluate locally. The text is not sent to any network service. | |
| context | No | Writing context. Technical mode suppresses patterns common in code-adjacent prose; defaults to general. | general |
Output Schema
ParametersJSON Schema
| Name | Required | Description |
|---|---|---|
| label | Yes | |
| score | Yes | |
| issues | Yes | |
| context | Yes | |
| scorable | Yes | |
| truncated | Yes | |
| confidence | Yes | |
| highlights | Yes | |
| statistics | Yes | |
| word_count | Yes | |
| issue_count | Yes | |
| probabilities | Yes | |
| classification | Yes | |
| unscored_reason | Yes |
TDQS
A4.7/5.0
Behavior5/5
Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?
Annotations already declare readOnlyHint, idempotentHint, and destructiveHint, and the description adds valuable behavior beyond that: the audit runs locally, is deterministic, returns up to 100 flagged patterns and 100 highlighted sentence regions, and includes truncation counts. It also disclaims that the result is heuristic rather than proof, which is essential context for the agent's interpretation.
Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.
Conciseness5/5
Is the description appropriately sized, front-loaded, and free of redundancy?
Three sentences, each earning its place: one identifies the action and method, one lists the return bounds and truncation behavior, and one sets expectations and routes to the sibling. The most decision-relevant facts are front-loaded.
Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.
Completeness5/5
Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?
Given the output schema exists, the annotations cover safety, and the description covers local processing, deterministic behavior, response bounds, truncation counts, and the heuristic caveat, an agent has everything it needs to select and invoke the tool correctly. The only thing left to the schema is the context enum, which is appropriately delegated.
Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.
Parameters3/5
Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?
Schema description coverage is 100%, so the text and context parameters are already documented. The description adds no extra parameter semantics, but it doesn't need to because the schema carries that weight. Baseline 3 is appropriate.
Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.
Purpose5/5
Does the description clearly state what the tool does and how it differs from similar tools?
The description states a specific action ('Audit text'), names the concrete detector ('the deterministic Avoid AI Writing detector'), and distinguishes this tool from its sibling score_text by describing the richer audit output. An agent can tell exactly what this tool does and how it differs from the sibling without opening the schema.
Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.
Usage Guidelines5/5
Does the description explain when to use this tool, when not to, or what alternatives exist?
The description gives clear usage context: it is a local, deterministic, heuristic writing audit and not proof of authorship. It explicitly says to use score_text when a compact result is needed, naming the alternative and the condition that selects it.
Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.
score_textScore text for AI-writing patternsARead-onlyIdempotent
Score text locally with the deterministic Avoid AI Writing detector. Returns a compact 0-100 pattern score, classification, confidence, and counts. This is a heuristic signal, not proof of authorship; use audit_text when individual findings are needed.
ParametersJSON Schema
| Name | Required | Description | Default |
|---|---|---|---|
| text | Yes | Text to evaluate locally. The text is not sent to any network service. | |
| context | No | Writing context. Technical mode suppresses patterns common in code-adjacent prose; defaults to general. | general |
Output Schema
ParametersJSON Schema
| Name | Required | Description |
|---|---|---|
| label | Yes | |
| score | Yes | |
| context | Yes | |
| scorable | Yes | |
| confidence | Yes | |
| word_count | Yes | |
| issue_count | Yes | |
| probabilities | Yes | |
| classification | Yes | |
| unscored_reason | Yes |
TDQS
A4.7/5.0
Behavior5/5
Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?
Annotations already cover read-only, idempotent, and non-destructive hints. The description adds critical behavioral traits: the detector is deterministic, it is a heuristic signal not proof of authorship, and it operates locally. These go beyond the annotations and help calibrate trust in the output.
Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.
Conciseness5/5
Is the description appropriately sized, front-loaded, and free of redundancy?
Two sentences with no filler. The action and scope are front-loaded, the caveat is stated compactly, and the alternative is offered in the same breath. Every word earns its place.
Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.
Completeness5/5
Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?
The tool has a rich output schema, complete parameter documentation, and strong annotations. The description covers the operation's purpose, the caveat about heuristic results, and the routing to audit_text. Nothing an agent needs to call it correctly is missing.
Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.
Parameters3/5
Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?
Schema description coverage is 100% – both parameters have clear descriptions (text is local, context has default and behavior). The description itself does not add parameter-specific semantics beyond the schema, so the baseline of 3 is appropriate.
Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.
Purpose5/5
Does the description clearly state what the tool does and how it differs from similar tools?
The description states a specific verb ('Score'), a resource ('text'), and the exact detector ('Avoid AI Writing'). It also names the sibling tool 'audit_text' and differentiates by saying when audit_text is needed, so it clearly distinguishes from the alternative.
Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.
Usage Guidelines5/5
Does the description explain when to use this tool, when not to, or what alternatives exist?
The description explicitly says to 'use audit_text when individual findings are needed', implying this tool is for a compact score. It also notes the operation is local and deterministic, giving clear context for when to choose this tool.
Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.
Tool Schema Changelog
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
1 tool update
- Changed
audit_text4 fields changed- added
Output schema / properties / highlights / maxItemsAdded value: +100 - added
Output schema / properties / issues / maxItemsAdded value: +100 - added
Output schema / properties / truncatedAdded value: +{ + "additionalProperties": false, + "properties": { + "highlights": { + "maximum": 9007199254740991, + "minimum": 0, + "type": "integer" + }, + "issues": { + "maximum": 9007199254740991, + "minimum": 0, + "type": "integer" + } + }, + "required": [ + "issues", + "highlights" + ], + "type": "object" +} - changed
Output schema / requiredPrevious value: -[ - "score", - "label", - "scorable", - "unscored_reason", - "classification", - "confidence", - "probabilities", - "word_count", - "issue_count", - "context", - "issues", - "statistics", - "highlights" -]New value: +[ + "score", + "label", + "scorable", + "unscored_reason", + "classification", + "confidence", + "probabilities", + "word_count", + "issue_count", + "context", + "issues", + "statistics", + "highlights", + "truncated" +]
2 tool updates
v0.1.0- First observed
audit_text - First observed
score_text
TDQS
A4.7/5.0
Scored across 2 tools
Disambiguation5/5
score_text and audit_text have clearly separated purposes: one returns a compact classification while the other returns detailed flagged patterns and highlighted regions. The cross-references in the descriptions remove any real ambiguity about which tool to use.
Naming Consistency5/5
Both tool names follow the same verb_noun pattern with lowercase snake_case and concise action words. This is fully consistent and predictable.
Tool Count4/5
Two tools is slightly below the typical 3-15 range, but for a narrowly scoped local text detector the pair covers both a compact score and a detailed audit. Each tool has a distinct, useful role, so the small count feels intentional rather than inadequate.
Completeness5/5
The server's apparent domain is deterministic AI-writing detection on text, and the two tools cover the full user journey: getting a quick judgment or drilling into specific flagged patterns and highlighted regions. There are no obvious missing operations for this stated purpose.
Maintenance
ActivityMaintained
ResponsivenessNo issues
Related MCP Connectors
Find AI-isms with evidence and fingerprint a writing voice from samples. 3 of 5 free.
Deterministic prompt-injection detector; signed, offline-verifiable verdicts. Not an LLM.
Six tools for SEO and AI-readability audits. 91 checks, 11 score modules.
49 deterministic tools for text integrity, agent control, and contextual quality evidence.
Related MCP Servers
- AlicenseAqualityBmaintenanceA private, open-source AI-text checker. Get a read on whether text looks AI-written, the exact AI-tell spans to fix, a reuse check, and a grammar pass.4MIT
- AlicenseAqualityAmaintenanceBilingual (EN/ES) AI-writing detection that shows the evidence instead of a percentage: named tells with line and column, hidden-character inspection, and citation cross-checking against a document's own bibliography. Seven of its nine tools run entirely locally and never touch the network.1023MIT
- AlicenseNot gradedqualityCmaintenanceai-detect is a free AI text detector that runs entirely on your own hardware. It scores writing sentence by sentence using desklib/ai-text-detector-v1.01, a 304-million-parameter DeBERTa-v3-large model that ranks first on the RAID detection benchmark. No API key, no per-word pricing and no upload: after a one-off 1.7 GB model download it just works offline. by Houtini4MIT