Caelara Labs Agent Safety tools over remote MCP. Pay-per-call via x402 USDC or API key.
Maintain this server? Claim it
Call this server's tools from your browser. Nothing to install.
Requests go through mcptry because browsers can't call most MCP servers directly. Header values are sent to the server for that request only and are never stored or logged.
Tool list from our last check. Connect live to refresh it.
Scan untrusted input: injection risk + PII redaction + sanitized text + allow/review/block verdict
Untrusted input text
Redact PII in sanitized output (default true)
This tool isn't marked read-only, so it may change something on the server or in an account you connected.
Pick your client. Values in capitals, like YOUR_TOKEN, are yours to fill in.
Run in a terminal
claude mcp add --transport http agent https://agent.caelaralabs.com/mcp
Scan untrusted input: injection risk + PII redaction + sanitized text + allow/review/block verdict
| Argument | Type | Description |
|---|---|---|
| text* | string | Untrusted input text |
| redact_pii | boolean | Redact PII in sanitized output (default true) |
{
"type": "object",
"required": [
"text"
],
"properties": {
"text": {
"type": "string",
"description": "Untrusted input text"
},
"redact_pii": {
"type": "boolean",
"description": "Redact PII in sanitized output (default true)"
}
}
}Validate LLM output: repair JSON, enforce schema, check grounding against a source, extract citations
| Argument | Type | Description |
|---|---|---|
| output* | string | Raw LLM output |
| schema | object | JSON Schema subset to enforce |
| source | string | Source text for grounding check |
{
"type": "object",
"required": [
"output"
],
"properties": {
"output": {
"type": "string",
"description": "Raw LLM output"
},
"schema": {
"type": "object",
"description": "JSON Schema subset to enforce"
},
"source": {
"type": "string",
"description": "Source text for grounding check"
}
}
}Extract URLs, DOIs, arXiv ids and author-year citations
| Argument | Type | Description |
|---|---|---|
| text* | string | Text containing citations |
{
"type": "object",
"required": [
"text"
],
"properties": {
"text": {
"type": "string",
"description": "Text containing citations"
}
}
}Split text into token-bounded overlapping chunks
| Argument | Type | Description |
|---|---|---|
| text* | string | Text to chunk |
| overlap | integer | Overlap tokens (default 50) |
| max_tokens | integer | Max tokens per chunk (default 500) |
{
"type": "object",
"required": [
"text"
],
"properties": {
"text": {
"type": "string",
"description": "Text to chunk"
},
"overlap": {
"type": "integer",
"description": "Overlap tokens (default 50)"
},
"max_tokens": {
"type": "integer",
"description": "Max tokens per chunk (default 500)"
}
}
}Flag answer sentences unsupported by the source text
| Argument | Type | Description |
|---|---|---|
| answer* | string | Model answer to check |
| source* | string | Source / ground-truth text |
{
"type": "object",
"required": [
"source",
"answer"
],
"properties": {
"answer": {
"type": "string",
"description": "Model answer to check"
},
"source": {
"type": "string",
"description": "Source / ground-truth text"
}
}
}Repair malformed JSON from LLM output
| Argument | Type | Description |
|---|---|---|
| text* | string | Malformed JSON text |
{
"type": "object",
"required": [
"text"
],
"properties": {
"text": {
"type": "string",
"description": "Malformed JSON text"
}
}
}Redact emails, phones, SSNs, cards (Luhn), IPs, API keys and JWTs
| Argument | Type | Description |
|---|---|---|
| text* | string | Text to redact |
{
"type": "object",
"required": [
"text"
],
"properties": {
"text": {
"type": "string",
"description": "Text to redact"
}
}
}Score text for prompt-injection patterns
| Argument | Type | Description |
|---|---|---|
| text* | string | Text to scan |
{
"type": "object",
"required": [
"text"
],
"properties": {
"text": {
"type": "string",
"description": "Text to scan"
}
}
}Validate and coerce data to a JSON Schema
| Argument | Type | Description |
|---|---|---|
| data* | object | Data to enforce |
| schema* | object | JSON Schema subset |
{
"type": "object",
"required": [
"data",
"schema"
],
"properties": {
"data": {
"type": "object",
"description": "Data to enforce"
},
"schema": {
"type": "object",
"description": "JSON Schema subset"
}
}
}Compress text to a target ratio of its original length
| Argument | Type | Description |
|---|---|---|
| text* | string | Text to compress |
| ratio | number | Target ratio 0.1-0.9 (default 0.5) |
{
"type": "object",
"required": [
"text"
],
"properties": {
"text": {
"type": "string",
"description": "Text to compress"
},
"ratio": {
"type": "number",
"description": "Target ratio 0.1-0.9 (default 0.5)"
}
}
}We read every tool's name, description and inputs the way the model will, looking for instructions hidden from you. A clean result means no warning signs, not a guarantee. Scanned Nov 19, 2026.
No readOnlyHint / destructiveHint annotations, so clients can't tell a read from a write without asking.
Fix: Add annotations to each tool, e.g. { readOnlyHint: true } for tools that only read.