Ein MCP-Server, der jedem KI-Agenten die Möglichkeit gibt, einen gerade erzeugten Regex vor der Auslieferung zu überprüfen – sowohl das, was er tatsächlich matcht, als auch die Frage, ob es sicher ist, ihn gegen nicht vertrauenswürdige Eingaben auszuführen. Vollständig lokal: kein API-Schlüssel, kein Netzwerkaufruf, keine weiteren Abhängigkeiten außer dem MCP SDK und Zod.
Reguläre Ausdrücke sind notorisch dafür bekannt, dass sich in ihnen leicht ein Bug einschleicht, der in jedem Beispiel, das Sie zufällig testen, völlig in Ordnung aussieht. Zwei Fehlermodi sind besonders häufig und mit bloßem Auge leicht zu übersehen:
Es matcht nicht das, was Sie denken, dass es matcht. explain_regex wandelt das Muster in einen echten Syntaxbaum um und beschreibt ihn in einfachem Englisch, sodass die Frage „Erfordert das wirklich mindestens eine Ziffer?“ schnell eine Antwort erhält, die nicht davon abhängt, dass Sie Ihrer eigenen Lesart von verschachtelten Klammern vertrauen.
Es ist ein Denial-of-Service-Vektor. Ein Regex mit verschachtelten Quantifizierern ((a+)+) oder mehrdeutiger Alternation innerhalb einer wiederholten Gruppe ((a|a)+) kann eine Backtracking-Engine dazu bringen, bei einer manipulierten (oder sogar versehentlichen) nicht übereinstimmenden Eingabe exponentielle Zeit zu benötigen – das ist ReDoS, eine reale und wiederholt ausgenutzte Schwachstellenklasse und ein durchaus möglicher Fehler in jedem Regex, den ein Agent schreibt, ohne ihn gegen böswillige Eingaben zu testen. check_redos_risk kennzeichnet die strukturelle Form, ohne das Muster jemals auszuführen – es lässt sich sicher auf nicht vertrauenswürdige oder absichtlich bösartige Regex-Quelltexte anwenden.
Related MCP server: cron-mcp
explain_regexParst ein Muster in einen AST und gibt eine Beschreibung in einfachem Englisch davon zurück, was es matcht.
check_redos_riskAnalysiert die Struktur eines Musters statisch auf verschachtelte Quantifizierer und mehrdeutige Alternation innerhalb einer wiederholten Gruppe – die beiden klassischen Ursachen für katastrophales Backtracking. Gibt "safe", "high" oder "critical" zurück, jeweils mit einem konkreten Befund für jedes gefundene Problem.
Beide Werkzeuge teilen sich einen Parser (src/services/parser.ts): einen echten rekursiv-absteigenden Regex-Parser (Literale, Zeichenklassen, Kurzzeichenklassen, Anker, erfassende/nicht erfassende/benannte Gruppen, Lookaround, Alternation, Quantifizierer, Rückverweise), keine Ansammlung von String-Matching-Heuristiken gegen den rohen Quelltext des Musters.
Dies ist eine heuristische Strukturprüfung, kein formaler Verifizierer – check_redos_risk kann Ihnen sagen, dass ein Muster die typische Form der exponentiellen Explosion hat; es kann nicht beweisen, dass ein Muster auf allen Eingaben schnell ist, und es gibt ReDoS-Muster außerhalb der beiden Formen, die es derzeit erkennt. Behandeln Sie ein "safe"-Ergebnis als „nichts Offensichtliches gefunden“, nicht als Garantie.
git clone https://github.com/wedo911/regexguard-mcp-server.git
cd regexguard-mcp-server
npm install
npm run buildFügen Sie es zur Konfiguration Ihres MCP-Clients hinzu (z. B. claude_desktop_config.json oder einem .mcp.json eines Projekts für Claude Code):
{
"mcpServers": {
"regexguard": {
"command": "node",
"args": ["/absolute/path/to/regexguard-mcp-server/dist/index.js"]
}
}
}npm run build
node --test tests/parser.test.mjs tests/explain.test.mjs tests/redosCheck.test.mjs44 Tests decken die Parser-Grammatik, die Erklärungsausgabe und sowohl echte Positivfälle ((a+)+, (a*)*, (a|a)+, (a|ab)+, Muster, die in nicht erfassenden Gruppen verschachtelt sind) als auch echte Negativfälle ((cat|dog)+, ein realistisches Benutzernamen-Muster, ein realistisches E-Mail-Muster, benachbarte – nicht verschachtelte – Wiederholungen) für den ReDoS-Check ab, sodass die False-Positive-Rate bei gewöhnlichen Mustern eine getestete Eigenschaft und keine Hoffnung ist.
npx @modelcontextprotocol/inspector --cli node dist/index.js \
--method tools/call --tool-name check_redos_risk \
--tool-arg pattern='^(([a-zA-Z0-9])+([\.-]?([a-zA-Z0-9])+)*)$'MIT – siehe LICENSE.
Statically analyze a regex's structure for the two classic causes of catastrophic backtracking: nested quantifiers (e.g. "(a+)+") and ambiguous alternation inside a repeated group (e.g. "(a|a)+"). Both can make a backtracking regex engine take exponential time on a crafted or even accidental non-matching input -- a real, exploitable denial-of-service vector, and a common defect in regexes generated without testing against adversarial input.
This tool NEVER executes the pattern -- it only parses and inspects the pattern's source structure, so it's safe to run on untrusted or deliberately malicious patterns without risk of hanging.
Args:
pattern (string, 1-1000 chars): the regex source, without delimiters.
Returns: For JSON format: { "parsed": boolean, "error": string | null, "risk": "safe" | "high" | "critical", "findings": [ { "category": "nested_quantifier" | "ambiguous_alternation", "severity": "high" | "critical", "message": string } ] }
Examples:
Use when: "is this regex I just generated safe to run against user input?" -> pass the pattern before using it
Use when: reviewing a regex from an untrusted source before adding it to a codebase
Don't use when: you need proof a regex is fast on all inputs -- this is a heuristic structural check (no false negatives are guaranteed to be caught), not a formal verifier
Error Handling:
Returns an error (not an exception) if the pattern doesn't parse.
| Name | Required | Description | Default |
|---|---|---|---|
| pattern | Yes | The regex source, without the surrounding slashes, e.g. "^(a+)+$". |
Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?
Annotations already indicate read-only, idempotent, non-destructive. The description adds critical behavioral context: the tool never executes the pattern, is safe on untrusted or malicious input, returns errors rather than exceptions on parse failure, and is heuristic with possible missed risks. This is significant added transparency with no contradiction.
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 with headings for purpose, usage, arguments, return format, examples, and error handling. Every section contributes useful operational information and the purpose is front-loaded within the first sentence.
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?
Since no output schema exists, the description provides the full JSON return structure, error behavior, and usage boundaries, making the tool actionable. Minor ambiguity about regex flavor is acceptable given the structural check nature, and the description is otherwise complete.
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%: the pattern parameter is fully described with type, min/max length, and an example in the schema. The description repeats the 'without delimiters' point but adds no new semantic meaning beyond what the schema already provides. Baseline 3 applies.
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 verb and resource: statically analyzing a regex's structure for the two classic causes of catastrophic backtracking (nested quantifiers and ambiguous alternation). This clearly distinguishes it from a generic regex explainer like the sibling explain_regex, even without naming it.
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?
Provides explicit 'Use when' scenarios (generated regex before use, reviewing untrusted regex) and a clear 'Don't use when' caution that it is a heuristic structural check, not a formal verifier. This gives the agent clear selection criteria and exclusions.
Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.
Parse a regular expression into a real syntax tree and describe in plain English what it matches. Use this to sanity-check a regex you (or someone else) wrote actually does what you intended, before shipping it.
Args:
pattern (string, 1-1000 chars): the regex source, without delimiters (no leading/trailing "/").
Returns: For JSON format: { "explanation": string, "parsed": boolean, "error": string | null } explanation is empty and error is set if the pattern could not be parsed (e.g. unbalanced parentheses, unsupported syntax).
Examples:
Use when: "does this regex actually require the string to start with a letter?" -> pass the pattern, read the explanation
Don't use when: you need to actually run the regex against text -- this tool never executes the pattern, only analyzes its source
Error Handling:
Returns an error (not an exception) if the pattern doesn't parse, with the parser's best guess at where the problem is.
| Name | Required | Description | Default |
|---|---|---|---|
| pattern | Yes | The regex source, without the surrounding slashes, e.g. "^[a-z0-9_]{3,16}$". |
Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?
Beyond the annotations, the description discloses important behavioral traits: the tool never executes the pattern, returns a parsed boolean, returns an error rather than throwing an exception on unparseable input, and provides the parser's best guess at the problem. This adds real context beyond the readOnly/idempotent hints and contains no contradiction.
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 with Args, Returns, Examples, and Error Handling sections, and it front-loads the core purpose. It is slightly redundant with the schema in the Args section, but the overall organization makes it easy for an agent to parse and apply.
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 only one parameter and no output schema, the description fully compensates by specifying the exact JSON return shape, the meaning of each field, error behavior, and a concrete usage example. An agent has enough information to call the tool correctly and interpret its result.
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 description coverage is 100%, and the description's Args section essentially repeats the schema: pattern string, 1-1000 chars, no delimiters. It adds no new semantic information beyond what the input schema already provides, so the baseline score of 3 is appropriate.
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 verb and resource: 'Parse a regular expression into a real syntax tree and describe in plain English what it matches.' It also clearly differentiates the tool from running a regex by explicitly saying it 'never executes the pattern, only analyzes its source,' which distinguishes it from execution-oriented tools.
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 guidance: 'Use this to sanity-check a regex... before shipping it' and includes both a 'Use when' example and a 'Don't use when' exclusion. It does not explicitly name check_redos_risk as the alternative for security risk assessment, so it misses the full when/when-not/alternatives pattern, but the context is otherwise clear.
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.
v1.0.0check_redos_riskexplain_regexScored across 2 tools
The two tools have clearly distinct purposes: explain_regex describes what a pattern matches, while check_redos_risk analyzes vulnerability to catastrophic backtracking. They share input format but have no functional overlap, making misselection unlikely.
Both tools follow a consistent verb_noun snake_case pattern (explain_regex and check_redos_risk). The naming is predictable and aligns with the domain.
With only two tools, the server is minimally scoped, but for a dedicated regex-analysis utility this is reasonable and each tool addresses a core need. Slightly under typical range but appropriate for the narrow focus.
The server covers the essential aspects of regex analysis: comprehension (explain_regex) and security risk (check_redos_risk). Syntax validation is implicitly handled via parse errors. There are no obvious missing operations within the stated purpose.
Explain a regex in plain English and detect catastrophic backtracking risk.
Security scanner for AI agents: prompt injection, leaked secrets, PII, and SSRF-risk URL detection.
The WAF for agents. Pattern-based + heuristic firewall scans prompts, RAG documents, tool argume...
Detects catastrophic-backtracking (ReDoS) regex risk via AST parsing, with JS-safe rewrites.