Tool Chronicle

On the record

webmcp.patronus.studio

Read 1 time since 11 Oct 2026. Last read 11 Oct 2026.

6

tools offered to AI agents, on the record.

https://webmcp.patronus.studio/

Last change
None yetUnchanged since first read.
Declared read-only
4 of 6
llms.txt
None

Tools

6 live
  1. get_scan_detailsReads only

    Open Details and return the latest completed scan of the requested kind in this page session, without another network request or scan. Use after scan_mcp_server or scan_url (or a scan through the UI). kind="mcp" returns a JSON string: { url, inspection: { serverName, protocolVersion, latency_ms, entries: [{ kind, name, scanText }] }, findings: [{ kind, name, scanText, result: { verdict, confidence

    First seen 11 Oct 2026. Unchanged since.

  2. scan_fileCan change things

    Scan a file previously uploaded with upload_file or the Patronus repository plugin. Pass only its opaque file_id, never file content, a local path or a filename. Returns a JSON string: { filename: string, verdict: "attack" | "benign", confidence: number, evidence: Array<{ start: number, end: number, text: string }>, latency_ms: number }. filename comes from upload metadata; evidence contains at mo

    First seen 11 Oct 2026. Unchanged since.

  3. scan_mcp_serverReads only

    Scan descriptions from a public HTTPS MCP endpoint, not an HTML page (use scan_url for pages). Inspects server instructions and advertised tools, prompts and resources without executing capabilities. Checks at most 25 entries, first 1,000 characters each, without listing pagination. Returns a JSON string summary: { server: string, objects: number, attacks: number, latency_ms: number }. objects is

    First seen 11 Oct 2026. Unchanged since.

  4. scan_textReads only

    Scan supplied text (1–1,000 characters) for prompt injection with Patronus Ark. Returns a JSON string: { verdict: "attack" | "benign", confidence: number, evidence: Array<{ start?: number, end?: number, text?: string }>, latency_ms: number }. Confidence is the classifier confidence, not a general safety score; evidence contains at most 5 spans. Benign means no prompt injection detected in this tex

    First seen 11 Oct 2026. Unchanged since.

  5. scan_urlReads only

    Scan the complete normalized static body of one public HTTPS HTML page for prompt injection. Does not crawl links or execute JavaScript; use scan_mcp_server for MCP endpoint metadata. Returns a JSON string with safety_status ("attack" | "unknown" | "benign"), verdict ({ class_name, confidence, ... }), evidence_spans, coverage ({ included, excluded, html_bytes, extracted_characters, segments_scanne

    First seen 11 Oct 2026. Unchanged since.

  6. upload_fileCan change things

    Upload a local document without sending its bytes through tool arguments or model context. First open the Document tab and attach the authorized local file through the browser file chooser (Choose or drop a document). Then call upload_file with {}. The browser sends the selected File directly to Patronus. Accepts supported UTF-8 source/text files up to 20 KiB. Returns a JSON string: { file_id: str

    First seen 11 Oct 2026. Unchanged since.

Run this site? This page shows what your site told our crawler, ToolChronicleBot. To be left out, disallow it in robots.txt or write to hello@toolchronicle.com.