# GrindFree Agent Security (npm · @grindfree/agent-security-mcp)

Pay-per-call agent security (x402): injection, tools, shell/SQL, memory, leaks, docs, safe reads

- Trust score: 67/100 (medium)
- Change this week: +3
- Registry status: active
- Liveness: live
- Owner verified: no
- Last scored: 2026-11-21

## Components

- npm · `@grindfree/agent-security-mcp`: 67/100 (this document), [markdown](https://verifymcp.io/servers/ronin93-grindfree-agent-security/grindfree-agent-security-mcp.md), [page](https://verifymcp.io/servers/ronin93-grindfree-agent-security/grindfree-agent-security-mcp)

## Channel facts

- Registry: `npm`
- Package: `@grindfree/agent-security-mcp`
- Version: `0.8.0`
- Transport: `stdio`

## Trust breakdown

How this component scores in each security and reliability category. Every signal is checked automatically from public evidence about the published package, including repeated runs of it in an isolated sandbox, and we only credit what we can confirm. Scores are 0–100 per category. Scoring method: https://verifymcp.io/docs/scoring (what has changed: https://verifymcp.io/docs/scoring/changelog)

Scored 2026-11-21.

- **Supply Chain Security**: 81/100
  - No malware found by supply-chain analysis.
  - CVE check failed: a known high-severity CVE affects zod 3.25.76, a direct dependency. A fixed version is available.
  - No install/post-install scripts declared.
  - 31 of 112 dependencies flagged as unhealthy.
- **Provenance & Transparency**: 45/100
  - Source repository is publicly reachable at the declared URL.
  - Provenance check failed: no build-provenance attestation is published.
  - Clear OSI-approved license (MIT).
  - Actively maintained (last published 0 days ago).
  - Disclosure check failed: no security disclosure policy was found in the source repository.
- **Schema Quality & AI Usability**: 61/100
  - AI-judged instruction clarity (good).
  - Context-footprint check failed: tool/resource definitions use about 3393 tokens (~161/item across 21 items; 21 tools + 0 resources), over budget; trim descriptions and params.
  - Usage-examples check failed: none of the tools include examples.
- **Stability & Change Management**: 40/100
  - Stability observed for 12 of 30 days with no destabilising changes; credit accrues until the full window elapses.
- **Tool Coverage**: 88/100
  - 100% of tools have a non-trivial description (not blank, and not just the tool's name).
  - 65% of tool parameters carry a description.
- **Tool Safety**: 100/100
  - No prompt-injection markers were found in the server instructions, tool names or descriptions we captured.
  - We read all 21 captured tool definition(s), and no name or description among them implies an irreversible operation.
  - An AI judge read all 21 captured unit(s) of tool text and found none that tries to manipulate the model reading it.
- **Capabilities**: 100/100
  - Implements a supported MCP spec version (2025-11-25); the latest is 2026-07-28.

## Install

### How do I install the GrindFree Agent Security MCP server?

GrindFree Agent Security runs locally as an npm package, launched with npx -y @grindfree/agent-security-mcp. Ready-made configuration for Claude, Cursor, VS Code, Codex and 5 more is on this page, copied from each client's own documentation.

### Claude

```bash
claude mcp add ronin93-grindfree-agent-security -- npx -y @grindfree/agent-security-mcp
```

### Cursor

```json
{
  "mcpServers": {
    "ronin93-grindfree-agent-security": {
      "command": "npx",
      "args": [
        "-y",
        "@grindfree/agent-security-mcp"
      ]
    }
  }
}
```

### VS Code

```json
{
  "servers": {
    "ronin93-grindfree-agent-security": {
      "command": "npx",
      "args": [
        "-y",
        "@grindfree/agent-security-mcp"
      ]
    }
  }
}
```

### Codex

```bash
codex mcp add ronin93-grindfree-agent-security -- npx -y @grindfree/agent-security-mcp
```

### opencode

```json
{
  "$schema": "https://opencode.ai/config.json",
  "mcp": {
    "ronin93-grindfree-agent-security": {
      "type": "local",
      "command": [
        "npx",
        "-y",
        "@grindfree/agent-security-mcp"
      ],
      "enabled": true
    }
  }
}
```

### OpenClaw

```bash
openclaw mcp add ronin93-grindfree-agent-security --command npx --arg -y --arg @grindfree/agent-security-mcp
```

### Hermes

```yaml
mcp_servers:
  ronin93-grindfree-agent-security:
    command: "npx"
    args: ["-y", "@grindfree/agent-security-mcp"]
```

### Netclaw

```json
{
  "McpServers": {
    "ronin93-grindfree-agent-security": {
      "Transport": "stdio",
      "Command": "npx",
      "Arguments": [
        "-y",
        "@grindfree/agent-security-mcp"
      ]
    }
  }
}
```

### Vellum

```bash
assistant mcp add ronin93-grindfree-agent-security -t stdio -c npx -a -y @grindfree/agent-security-mcp
```

### Other

```json
{
  "mcpServers": {
    "ronin93-grindfree-agent-security": {
      "command": "npx",
      "args": [
        "-y",
        "@grindfree/agent-security-mcp"
      ]
    }
  }
}
```

## Changelog

Every change recorded for this component, newest first. Days that predate change tracking, or that we cannot explain, say so: "we were watching and nothing happened" and "we were not watching" are different claims.

### 2026-11-21 (score 67, +6)

- [security regression] Known CVEs: unverified → fail
- [functional improvement] Dependency health: unverified → 0.88

### 2026-11-20 (score 61, −5)

- [security regression] Known CVEs: fail → unverified
- [functional regression] Schema quality: 2926 → 3393
- [functional regression] Dependency health: 0.88 → unverified
- [functional] Package version: 0.7.0 → 0.8.0

### 2026-11-19 (score 66, +16)

- [security improvement] Malware scan: unverified → pass

### 2026-11-18 (score 50, −15)

- [security regression] Malware scan: pass → unverified
- [functional] Package version: 0.6.0 → 0.7.0

### 2026-11-17 (score 65, +15)

- [security improvement] Malware scan: unverified → pass
- [functional regression] Schema quality: 2245 → 2746
- [functional improvement] Tool coverage: 60% → 65%
- [functional] Package version: 0.5.0 → 0.6.0

### 2026-11-16 (score 50, −14)

- [security regression] Malware scan: pass → unverified
- [security regression] Stability: 0.20 → unverified
- [security regression] Tool safety: pass → unverified
- [functional regression] Schema quality: 1951 → 2245
- [functional regression] Capabilities: pass → unverified
- [functional regression] Tool coverage: 100 → unverified
- [functional improvement] Tool coverage: 55% → 60%
- [functional] First check of Schema quality: unverified
- [functional] Package version: 0.4.0 → 0.5.0

### 2026-11-14 (score 64, +1)

No change was recorded against any check on this day. Stability & Change Management went from 13 to 17. That category is still filling its 30-day observation window: 4 days of observed history at the previous scan, 5 at this one. The score rises as the window fills, whether or not the server changes.

### 2026-11-13 (score 63, +5)

- [security regression] Known CVEs: unverified → fail
- [functional improvement] Dependency health: unverified → 0.89

## MCP tools (21)

### `promptshield_screen` (~79 tokens)

Screen untrusted text/HTML for prompt-injection before the model reads it. Paid x402 $0.002 USDC on Base. Returns risk, findings, sanitized. On high risk, prefer sanitized or abort.

Input parameters:

- `content` (string, required): Untrusted text or HTML to screen
- `format` (string): Content format; default text

### `promptshield_clean` (~52 tokens)

Convert HTML to clean LLM-ready text (strips scripts/styles/hidden/zero-width). Paid x402 $0.002 USDC on Base.

Input parameters:

- `content` (string, required): HTML or text to clean

### `crawlsafe_check` (~114 tokens)

Check whether a URL may be crawled for your user-agent (robots.txt + X-Robots-Tag/meta). SSRF-guarded. Not a malware scanner. Paid x402 $0.001 USDC on Base.

Input parameters:

- `check_page` (boolean): Also fetch up to 64KB for X-Robots-Tag/meta (default true)
- `url` (string, required): http(s) URL to check
- `user_agent` (string): Crawler UA; first token matched against robots groups

### `toolguard_evaluate` (~142 tokens)

Pre-tool-call policy gate for HTTP/MCP outbound actions. Returns allow|review|deny with deterministic reasons. Caller MUST enforce. Not malware scanning or an exfil guarantee. Missing policy → review. Paid x402 $0.002 USDC on Base.

Input parameters:

- `arg_classes` (array)
- `destination` (object)
- `policy` (object): Owner policy; absent → review
- `policy_hash` (string)
- `rule_pack` (string)
- `secret_labels` (array)
- `taint` (array)
- `tool` (string, required): Tool name, e.g. http.fetch or mcp.call

### `commandguard_command` (~136 tokens)

Check an agent-generated shell command (bash/sh/zsh/fish/PowerShell/cmd) BEFORE executing it. Returns allow|warn|block with rule codes (rm -rf, curl|sh, force-push to protected branches, credential exfil, cloud/infra deletes, disk wipes). Caller MUST enforce: do not run on block. Paid x402 $0.001 USDC on Base.

Input parameters:

- `command` (string, required): Exact command line the agent intends to run
- `policy` (object): Optional owner policy, e.g. {mode:'strict', allow_rules:[...]}
- `shell` (string)

### `commandguard_sql` (~121 tokens)

Check agent-generated SQL BEFORE executing it: DROP/TRUNCATE, DELETE/UPDATE without WHERE, tautological WHERE, GRANT/role changes, file/command execution, writes in read_only mode, protected tables. Returns allow|warn|block. Caller MUST enforce. Paid x402 $0.001 USDC on Base.

Input parameters:

- `dialect` (string)
- `policy` (object): e.g. {read_only:true} or {protected_tables:['users'], allow_write_tables:['events']}
- `sql` (string, required): One or more SQL statements

### `memoryguard_screen` (~143 tokens)

Screen content BEFORE writing it to an agent's long-term memory: planted instructions, persistence/trigger attacks ('from now on…', 'whenever the user asks…'), exfil setups, secrets and PII. Source-aware (user vs tool/web/email). Returns allow|review|block plus a sanitized copy. Paid x402 $0.002 USDC on Base.

Input parameters:

- `content` (string): Single memory entry (or use entries)
- `entries` (array)
- `memory_type` (string)
- `policy` (object): Optional owner policy, e.g. {mode:'strict', allow_rules:[...]}
- `source` (string)

### `wireshield_scan` (~179 tokens)

Scan agent OUTPUT before it is rendered or sent: zero-click markdown-image exfil, data smuggled in URLs/DNS labels, deceptive links, leaked canary tokens, system-prompt leakage, hidden Unicode, active HTML, secrets and PII. Returns allow|review|block plus a sanitized copy. Paid x402 $0.002 USDC on Base.

Input parameters:

- `allowed_domains` (array): Hosts allowed to auto-load images
- `canaries` (array): Canary tokens planted in your system prompt
- `content` (string, required): Agent output about to be shown or sent
- `format` (string)
- `policy` (object): Optional owner policy, e.g. {mode:'strict', allow_rules:[...]}
- `system_prompt` (string): Optional; used only for verbatim-overlap detection, never stored

### `docshield_scan` (~215 tokens)

Scan a document BEFORE the agent reads it for hidden instructions: PDF (invisible/white/microscopic/off-page text, annotations, metadata, JavaScript), DOCX/XLSX/PPTX (hidden runs/sheets/rows/slides, speaker notes, comments, remote templates, DDE), HTML, Markdown. Returns allow|review|block and optional visible-only safe_text. Max 3 MB. Paid x402 $0.004 USDC on Base (unsupported/unparseable files are not charged).

Input parameters:

- `content` (string): HTML/Markdown/text (alternative to path)
- `content_base64` (string): File bytes, base64 (alternative to path)
- `filename` (string)
- `format` (string)
- `include_text` (boolean): Return visible-only safe_text
- `path` (string): Local file path (read by this MCP server on your machine)
- `policy` (object): Optional owner policy, e.g. {mode:'strict', allow_rules:[...]}

### `saferead_read` (~267 tokens)

Read a web page as clean Markdown for AI: fetch a URL and get LLM-ready Markdown (or text) with hidden content (display:none, font-size:0, off-screen, comments, invisible Unicode) and prompt-injection instructions removed or flagged. Returns title, content, removed/flagged segments with reasons, an allow|warn|block verdict, fetch metadata and the robots.txt decision. Honours robots.txt (pages disallowed for SafeReadBot return 403, not charged). SSRF-guarded. Use instead of a raw fetch for pages the agent did not write. Paid x402 $0.002 USDC on Base; only successful reads are charged.

Input parameters:

- `format` (string): Output format; default markdown
- `include_images` (boolean)
- `include_links` (boolean)
- `main_content` (boolean): Prefer <main>/<article>, drop nav/footer (default true)
- `max_chars` (integer): Max content characters (default 50000)
- `mode` (string): strip removes injected blocks (default); flag keeps and lists them
- `policy` (object): Optional owner policy, e.g. {mode:'strict', allow_rules:[...]}
- `url` (string, required): http(s) URL to read

### `saferead_fetch` (~180 tokens)

Fetch a web page URL as clean Markdown for AI (plain read): LLM-ready Markdown without scripts, styles, nav or footers, plus title, fetch metadata and the robots.txt decision. Honours robots.txt; SSRF-guarded. NO hidden-text removal or prompt-injection screening: for pages the agent did not write, use saferead_read instead. Paid x402 $0.001 USDC on Base; only successful reads are charged.

Input parameters:

- `format` (string): Output format; default markdown
- `include_images` (boolean)
- `include_links` (boolean)
- `main_content` (boolean): Prefer <main>/<article>, drop nav/footer (default true)
- `max_chars` (integer): Max content characters (default 50000)
- `url` (string, required): http(s) URL to fetch

### `saferead_read_batch` (~202 tokens)

Read up to 10 web pages as clean Markdown in one call: each URL returns LLM-ready Markdown with hidden text and prompt injections removed, plus a verdict and robots.txt decision. Honours robots.txt. All-or-nothing: if any URL fails (robots disallow, blocked host, fetch error) the call returns 424 with per-URL errors and nothing is charged. Paid x402 $0.002 USDC per URL on Base (e.g. 5 URLs = $0.010).

Input parameters:

- `format` (string): Output format; default markdown
- `main_content` (boolean): Prefer <main>/<article>, drop nav/footer (default true)
- `max_chars` (integer): Max content characters per page (default 20000)
- `mode` (string): strip removes injected blocks (default); flag keeps and lists them
- `urls` (array, required): 1-10 http(s) URLs (duplicates are merged)

### `saferead_watch` (~191 tokens)

Check web pages for changes on a schedule (daily digest, monitor): each URL is read with hidden text and prompt injections removed, compared with the last check under the same watch_id, and returned as new | changed | unchanged with only the added sections. Page text is never stored (only section hashes, kept 90 days). All-or-nothing; failures are not charged. Paid x402 $0.002 USDC per URL on Base (5 URLs = $0.010).

Input parameters:

- `format` (string)
- `max_chars` (integer)
- `return` (string): changes (default) returns only added sections; full returns the whole cleaned page
- `urls` (array, required): 1-10 http(s) URLs
- `watch_id` (string): Stable id for this watch list, e.g. 'my-daily-digest' (default: your paying wallet)

### `shipcheck_scan` (~173 tokens)

Scan an AI agent repo for security issues before merge or deploy: secrets, prompt injection and hidden Unicode in agent instruction files (AGENTS.md, CLAUDE.md, .cursorrules, SKILL.md), unpinned or poisoned MCP servers, inline MCP secrets, lethal-trifecta toolsets and risky CI/package/Docker commands. Returns pass/fail, findings with file and line, optional SARIF. Paid x402 $0.01 USDC on Base per run (up to 200 files, 1 MB).

Input parameters:

- `fail_on` (string): Fail on block findings (default) or review findings too
- `files` (array, required): Repo-relative path + text content
- `sarif` (boolean): Also return SARIF 2.1.0 for GitHub code scanning

### `grindfree_quote` (~103 tokens)

Free price quote before you spend (no payment, no wallet needed): give planned GrindFree calls and an optional budget, get the exact USDC total, whether it fits, and how many runs the budget buys. Every 402 also carries X-Price-USD; only HTTP 2xx is charged.

Input parameters:

- `budget_usd` (number)
- `calls` (array, required)
- `origin` (string): Which service to quote (default saferead)

### `docread_convert` (~327 tokens)

Convert PDF to Markdown for AI, also DOCX, XLSX, PPTX, HTML, Markdown or text, from a URL, a local file path or inline content. Returns Markdown (headings, lists, tables) built from visible content only, a report of what was removed (hidden/white/tiny/invisible text, hidden sheets or slides, prompt-injection lines), flagged items and an allow|warn|block verdict. Feed only `markdown` to the model. include_findings=true adds DocShield's full scan of every channel. Max 3 MB. Paid x402 $0.003 USDC on Base; only successful conversions are charged.

Input parameters:

- `content` (string): Inline HTML, Markdown or plain text
- `content_base64` (string): File bytes, base64 (alternative to url/path)
- `filename` (string): File name with extension, helps format detection
- `format` (string): Output format; default markdown
- `include_findings` (boolean): Also return docshield_findings: every channel scanned, what was removed and why
- `include_speaker_notes` (boolean): PPTX: include speaker notes
- `max_chars` (integer)
- `mode` (string): strip removes injected blocks (default); flag keeps and lists them
- `path` (string): Local file path (read by this MCP server on your machine)
- `policy` (object): Optional owner policy, e.g. {mode:'strict', allow_rules:[...]}
- `url` (string): http(s) URL of the document

### `docread_tables` (~249 tokens)

Extract tables from PDF, DOCX, XLSX, PPTX or HTML to typed JSON, from a URL, a local file path or inline content. Each visible table returns columns, column_types (number, currency, percent, date, boolean, string) and typed rows; as_records adds row objects. Hidden sheets, rows and text are never included; cells carrying prompt injections are replaced and listed in removed. No tables found returns 422, not charged. Max 3 MB. Paid x402 $0.004 USDC on Base; only successful extractions are charged.

Input parameters:

- `as_records` (boolean): Also return rows as objects keyed by column name
- `content` (string): Inline HTML or Markdown containing tables
- `content_base64` (string): File bytes, base64 (alternative to url/path)
- `filename` (string): File name with extension, helps format detection
- `header` (boolean): First row of each table is column names (default true)
- `max_tables` (integer)
- `path` (string): Local file path (read by this MCP server on your machine)
- `url` (string): http(s) URL of the document

### `trifectagate_check` (~190 tokens)

Session exfiltration gate (lethal trifecta): call BEFORE every outbound action (HTTP request, webhook, email, message, image URL, DNS, payment). Report what the session touched as events (private_data from inboxes/DBs/files; untrusted_content from web pages, docs, tool output, with SafeRead/PromptShield verdict if known). Returns allow | review | block with rule + reasons; review includes a ready AskHuman request. Taint is sticky per session_id. Paid x402 $0.002 USDC on Base (invalid input not charged).

Input parameters:

- `action` (object)
- `events` (array)
- `policy` (object)
- `session_facts` (object)
- `session_id` (string): Unguessable id for this agent session; reuse it on every call
- `session_ttl_seconds` (integer)

### `askhuman_request` (~216 tokens)

Human-in-the-loop approval: ask the agent's human owner to approve, deny or edit ONE proposed action before you run it (refunds, emails, deletes, deploys, payments, or any guard 'review' verdict). Pushes a one-tap review link via ntfy, Slack, Discord or a webhook (or returns it with link_only). Returns id, poll_token, wait_url and approval_url (give approval_url only to the reviewer). Then call askhuman_status until status != pending. Fails closed: no answer before timeout = denied. Paid x402 $0.03 USDC on Base (invalid input or a failed notification is not charged).

Input parameters:

- `action` (object, required)
- `allow_edit` (boolean)
- `callback_url` (string): https URL for an HMAC-signed decision callback
- `guard_decision` (object)
- `notify` (object, required): Exactly one channel
- `on_timeout` (string)
- `timeout_seconds` (integer): Default 3600

### `askhuman_status` (~82 tokens)

Get the human's decision for an AskHuman request (free). status: pending|approved|denied|edited|expired. Run the action only on approved (or edited, using edited_args). wait (1-25 s) long-polls.

Input parameters:

- `id` (string, required)
- `poll_token` (string, required)
- `wait` (integer)

### `grindfree_discovery` (~32 tokens)

Return machine-readable discovery URLs and unpaid 402 challenge shapes for all GrindFree agent-security APIs (no payment).

## Diagnostics

Captured diagnostic sections: Provenance, Vulnerabilities, Dependencies. The full working is on the page: https://verifymcp.io/servers/ronin93-grindfree-agent-security/grindfree-agent-security-mcp#diagnostics

## Score history

- 2026-11-21: 67
- 2026-11-20: 61
- 2026-11-19: 66
- 2026-11-18: 50
- 2026-11-17: 65
- 2026-11-16: 50
- 2026-11-15: 64
- 2026-11-14: 64
- 2026-11-13: 63
- 2026-11-12: 58
- 2026-11-11: 64
- 2026-11-10: 63
- 2026-11-09: 63

## Common questions

### What is the GrindFree Agent Security MCP server?

GrindFree Agent Security is an MCP server listed in the public MCP registry as io.github.ronin93/grindfree-agent-security. Pay-per-call agent security (x402): injection, tools, shell/SQL, memory, leaks, docs, safe reads. This page covers its npm package (@grindfree/agent-security-mcp).

### Is the GrindFree Agent Security MCP server safe to use?

GrindFree Agent Security scores 67 out of 100 on VerifyMCP. We recorded 1 known advisory against it as of 21 November 2026. It declares no install or post-install scripts. That is a record of what we were able to check automatically, not an endorsement. The category breakdown on this page shows every signal behind the number, including the ones we could not confirm.

### What tools does the GrindFree Agent Security MCP server expose?

GrindFree Agent Security exposes 21 tools: promptshield_screen, promptshield_clean, crawlsafe_check, toolguard_evaluate, commandguard_command, and 16 more. Their descriptions and schemas cost roughly 3,393 tokens of context every time the server is loaded.

### Is the GrindFree Agent Security MCP server still maintained?

GrindFree Agent Security is still listed as active in the MCP registry. We last reached this channel on 21 November 2026. Those dates come from our own scans of the registry and the channel itself, not from anything the publisher announced.

### What licence is the GrindFree Agent Security MCP server under?

GrindFree Agent Security declares the MIT licence, which is OSI-approved. That covers the source only, and says nothing about the cost of any service it calls.

### What did VerifyMCP test on the GrindFree Agent Security MCP server?

VerifyMCP ran 16 checks on GrindFree Agent Security, covering Malware scan, Known CVEs, Install scripts, Dependency health and 12 more. Endpoint Security and Transport & Reachability do not apply to an npm package, so they were not tested. The score does not measure everything, and the scoring documentation lists what it leaves out.

## Links

- npm package: https://www.npmjs.com/package/@grindfree/agent-security-mcp
- Socket report: https://socket.dev/npm/package/@grindfree/agent-security-mcp
- Repository: https://github.com/ronin93/grindfree-agent-security
- Changelog RSS feed: https://verifymcp.io/servers/ronin93-grindfree-agent-security/grindfree-agent-security-mcp.xml
- Changelog JSON feed: https://verifymcp.io/servers/ronin93-grindfree-agent-security/grindfree-agent-security-mcp/changelog.json
- Trust report (JSON): https://verifymcp.io/servers/ronin93-grindfree-agent-security/grindfree-agent-security-mcp.json
- HTML version of this page: https://verifymcp.io/servers/ronin93-grindfree-agent-security/grindfree-agent-security-mcp
