Vettory Vetted tools. Trusted results.
September 2025 · this already happened

A fake email tool quietly copied every message an agent sent.

It was identical to the real one — same name, same docs. Your best protection was knowing which of the two to trust.

That is what Vettory is: the trusted door where your agents pick up vetted tools — scored against a published rubric, and watched over time.

Same name. Same docs. One was a trap.
✓postmark-mcp

The official one — from ActiveCampaign. Vetted, cited, in the catalog.

▲postmark-mcp

The copycat — same name, published by a stranger. BCC'd every email to an attacker from v1.0.16 on.

Identical names. 1,643 downloads before anyone noticed. Your best defense was knowing the source. Source: The Hacker News ↗
22Tools scored
6Rubric dimensions
1Active warnings
3 Sep 2026Last re-check
The catalog

Every tool, scored and cited.

Every tool is scored against our six-part rubric and cited to its source. A green ✔ marks the ones a person has verified; the rest are researched and rated, with anything genuinely uncertain held back. Each row shows a verdict — trusted, caution, or warning — and the date it was last checked. Open one for watch-outs, history, and the source.

ToolVerdictRubricLast checked

Claude Native Scheduled Tasks — TRUSTED

Scheduling & Reminders · human-verified by a person at Vettory · last checked 28 Aug 2026

Built-in recurring tasks with push notifications. No code.

Best for: An agent that reaches out on a schedule with zero setup.

Watch out: Needs a paid Claude plan (Pro, Max, Team or Enterprise), with per-plan daily run limits; time can drift at daylight-saving.

Rubric: trust 5/5 · fit 4/5 · ease 5/5 · reliability 5/5 · reach 4/5 · cost 5/5

Sources: Anthropic — schedule recurring tasks

Twilio — TRUSTED

Scheduling & Reminders · rubric-scored; not personally verified · last checked 28 Aug 2026

The industry standard for real SMS, WhatsApp and voice.

Best for: When an agent needs to actually text a person, two-way, worldwide.

Watch out: The official Twilio MCP server is read-only — it searches and retrieves API documentation and cannot send a message, place a call, or touch your account. Sending requires your own API integration, or Twilio Labs' experimental twilio-alpha MCP server, which is not covered by Twilio Support and takes account credentials as a command-line argument. US texting needs carrier registration; publicly reported breaches (2022 phishing, 2024 Authy) have since been remediated.

Rubric: trust 5/5 · fit 5/5 · ease 2/5 · reliability 5/5 · reach 5/5 · cost 4/5

Sources: Twilio MCP server docs — read-only search/retrieve · Twilio Alpha MCP server (experimental, unsupported)

Resend — TRUSTED

Email · rubric-scored; not personally verified · last checked 3 Sep 2026

Modern, developer-first email with a hosted connector. Nothing to install.

Best for: A clean, modern, agent-ready way to send email.

Watch out: Email only; sending-domain DNS setup takes a few steps. Not HIPAA — Resend states plainly that it cannot sign a BAA. The free tier is 3,000 emails/month but capped at 100 a day, which is the limit that actually bites.

Rubric: trust 5/5 · fit 5/5 · ease 4/5 · reliability 4/5 · reach 3/5 · cost 5/5

Sources: Resend security — SOC 2 Type II, GDPR, not HIPAA · Resend pricing — free tier 3,000/mo, 100/day

Postmark — TRUSTED

Email · rubric-scored; not personally verified · last checked 28 Aug 2026

Famous for landing mail in the inbox — use only the official package.

Best for: When reaching the inbox reliably matters most.

Watch out: In Sept 2025 a malicious npm copy named `postmark-mcp` secretly BCC'd every email to an attacker — install only Postmark's own official server (github.com/ActiveCampaign/postmark-mcp). Free tier is 100 emails/mo.

Rubric: trust 5/5 · fit 5/5 · ease 4/5 · reliability 5/5 · reach 3/5 · cost 4/5

Sources: Snyk — malicious postmark-mcp on npm

Vendor page

SendGrid — TRUSTED

Email · rubric-scored; not personally verified · last checked 28 Aug 2026

Huge-scale, battle-tested — but no official agent connector.

Best for: High-volume sending on a proven platform.

Watch out: No official connector that can actually send — Twilio's official MCP server is read-only. sendgrid.com merged into twilio.com in Feb 2026 (branding and docs only; the API, SMTP endpoints and keys are unchanged). The permanent free tier was retired on 27 May 2025 and replaced by a 60-day trial (100 emails/day); not self-serve HIPAA; Twilio has past breaches, since remediated.

Rubric: trust 4/5 · fit 5/5 · ease 3/5 · reliability 5/5 · reach 3/5 · cost 3/5

Sources: Twilio changelog — SendGrid free plan changes · Twilio SendGrid Email API (sendgrid.com now redirects here) · Twilio MCP server — read-only, search and retrieve only

Vendor page

Slack — TRUSTED

Team Messaging · rubric-scored; not personally verified · last checked 28 Aug 2026

Post an agent's messages, alerts and approvals into channels and DMs.

Best for: Putting an agent's output where a team already works.

Watch out: Internal only; connector needs an internal or listed app.

Rubric: trust 5/5 · fit 4/5 · ease 4/5 · reliability 5/5 · reach 3/5 · cost 4/5

Sources: Slack MCP server docs

Tavily — TRUSTED

Live Web Search · rubric-scored; not personally verified · last checked 28 Aug 2026

The search layer most agent frameworks reach for first. Paste one key.

Best for: Searching the web with the least setup and a real free tier.

Watch out: Acquired by Nebius — announced 10 Feb 2026; the service continues under the Tavily brand. Draws on other providers, not its own index.

Rubric: trust 5/5 · fit 5/5 · ease 5/5 · reliability 4/5 · reach 4/5 · cost 5/5

Sources: Nebius — agreement to acquire Tavily, 10 Feb 2026

Vendor page

Exa — TRUSTED

Live Web Search · rubric-scored; not personally verified · last checked 3 Sep 2026

Its own index, finding things by meaning rather than keywords.

Best for: Finding the exact right thing by meaning, not keywords.

Watch out: Pricier per search; needs a little tuning. HIPAA and zero-data-retention are Enterprise-only — they require a sales conversation, not a signup.

Rubric: trust 5/5 · fit 5/5 · ease 4/5 · reliability 5/5 · reach 5/5 · cost 3/5

Sources: Exa enterprise security documentation · Exa Trust Center

Vendor page

Linkup — TRUSTED

Live Web Search · rubric-scored; not personally verified · last checked 3 Sep 2026

Privacy- and EU-minded search, easiest install, most generous free tier.

Best for: EU compliance, privacy, or a bigger free allowance.

Watch out: Youngest/smallest; a little vendor-longevity risk.

Rubric: trust 5/5 · fit 4/5 · ease 5/5 · reliability 3/5 · reach 4/5 · cost 5/5

Sources: Linkup security & privacy documentation

Vendor page

Perplexity — TRUSTED

Live Web Search · human-verified by a person at Vettory · last checked 28 Aug 2026

Returns ready-made cited answers, not just links.

Best for: Cited, ready-to-use answers with a documented compliance posture.

Watch out: DPA/BAA are Enterprise/sales; US-hosted; self-declared; token pricing. Disclosure: their Comet browser — a different product from this search API — had prompt-injection hijack disclosures in 2026 (Zenity Labs, LayerX), and an unproven class action over data sharing was filed in March 2026.

Rubric: trust 4/5 · fit 4/5 · ease 5/5 · reliability 5/5 · reach 4/5 · cost 3/5

Sources: Perplexity privacy & security docs · Zenity Labs — Comet agent-hijack disclosure

Mem0 — TRUSTED

Memory & Knowledge · rubric-scored; not personally verified · last checked 28 Aug 2026

A ready-to-use memory layer — no database to assemble yourself.

Best for: Giving an agent memory without wiring up infrastructure.

Watch out: Newer company. Their trust centre lists SOC 2 Type 1 — Type II is not shown — and HIPAA as a self-attestation rather than a certification.

Rubric: trust 4/5 · fit 5/5 · ease 5/5 · reliability 4/5 · reach 4/5 · cost 5/5

Sources: Mem0 Trust Center

Vendor page

Zep — TRUSTED

Memory & Knowledge · rubric-scored; not personally verified · last checked 28 Aug 2026

Tracks how facts change, so an agent knows what's true now.

Best for: When facts change over time (preferences, statuses).

Watch out: Small/early-stage; the self-hosted Community Edition was retired in 2025 — open-source work continues as Graphiti.

Rubric: trust 4/5 · fit 5/5 · ease 4/5 · reliability 3/5 · reach 4/5 · cost 4/5

Sources: Zep — end of Community Edition support

Vendor page

Pinecone — TRUSTED

Memory & Knowledge · rubric-scored; not personally verified · last checked 28 Aug 2026

The most turnkey vector database. Fully managed and certified.

Best for: A rock-solid managed vector store to build memory on.

Watch out: It's a database, not a memory layer; press reports in Aug 2025 said it was weighing a sale.

Rubric: trust 5/5 · fit 4/5 · ease 5/5 · reliability 5/5 · reach 5/5 · cost 4/5

Sources: Pinecone security · Reported sale talks (Aug 2025)

Weaviate — TRUSTED

Memory & Knowledge · rubric-scored; not personally verified · last checked 3 Sep 2026

Feature-rich open-source vector database, strong at scale.

Best for: A feature-rich open-source vector DB, strong at scale.

Watch out: The always-free tier is capped at one cluster and one collection; paid tiers step up steeply from there. HIPAA is Enterprise Cloud (AWS) only.

Rubric: trust 5/5 · fit 5/5 · ease 4/5 · reliability 5/5 · reach 5/5 · cost 4/5

Sources: Weaviate pricing — always-free tier · Weaviate security — SOC 2 Type II (Drata)

Vendor page

Chroma — WARNING

Memory & Knowledge · human-verified by a person at Vettory · last checked 3 Sep 2026

Confirmed maximum-severity remote-takeover flaw, still unpatched.

Best for: Local prototyping only — use the Rust version, never expose to the internet.

Watch out: NVD rates this pre-login remote-takeover bug 10.0/10 (CVE-2026-45829). Still unpatched — the advisory lists no patched version, and v1.5.9 remains the latest stable release. Use the Rust server; never expose the Python server to the internet.

Rubric: trust 1/5 · fit 4/5 · ease 5/5 · reliability 3/5 · reach 4/5 · cost 5/5

Sources: NVD — CVE-2026-45829 (CVSS 10.0) · GitHub advisory — no patched version

LlamaParse — TRUSTED

Documents & Files · rubric-scored; not personally verified · last checked 3 Sep 2026

Turns messy PDFs, scans and Office files into clean text.

Best for: Reading documents reliably with no infrastructure.

Watch out: No free self-host; a HIPAA BAA requires the Enterprise plan.

Rubric: trust 5/5 · fit 5/5 · ease 5/5 · reliability 4/5 · reach 5/5 · cost 4/5

Sources: LlamaCloud pricing & free credits · LlamaParse pricing

Vendor page

Reducto — TRUSTED

Documents & Files · rubric-scored; not personally verified · last checked 3 Sep 2026

Vision-first reader for the hardest files — tables, charts, handwriting.

Best for: Complex, high-stakes documents where accuracy is everything.

Watch out: No free self-host; accuracy scores are vendor-run.

Rubric: trust 5/5 · fit 5/5 · ease 4/5 · reliability 4/5 · reach 4/5 · cost 3/5

Sources: Reducto trust centre — ZDR, on-prem, air-gapped, HIPAA, SOC 2, BAA

Vendor page

Docling — TRUSTED

Documents & Files · rubric-scored; not personally verified · last checked 28 Aug 2026

Free, open-source, runs entirely on your own machine.

Best for: Parsing documents privately and for free (needs some code).

Watch out: Not turnkey; needs technical setup.

Rubric: trust 4/5 · fit 5/5 · ease 2/5 · reliability 5/5 · reach 5/5 · cost 5/5

Sources: Docling joins LF AI & Data · LF AI & Data — DocLang working group, June 2026

Vendor page

Unstructured — TRUSTED

Documents & Files · rubric-scored; not personally verified · last checked 3 Sep 2026

The broadest range of file types, with dozens of source connectors.

Best for: Pipelines ingesting many file types from many sources.

Watch out: HIPAA-aligned security is listed, but no Business Associate Agreement is documented — confirm one before putting PHI through it.

Rubric: trust 5/5 · fit 4/5 · ease 3/5 · reliability 4/5 · reach 5/5 · cost 5/5

Sources: Unstructured security & compliance overview

Vendor page

Stripe — TRUSTED

Payments · rubric-scored; not personally verified · last checked 28 Aug 2026

An official agent toolkit and MCP server, from a heavily-audited processor.

Best for: Handling real payments from a well-audited, widely-used processor.

Watch out: Primarily lets an agent RECEIVE money; agent-spending (Treasury) is still in preview. Toolkit package names are mid-restructure.

Rubric: trust 5/5 · fit 5/5 · ease 4/5 · reliability 5/5 · reach 5/5 · cost 5/5

Sources: Stripe — Agentic Treasury (preview, access by request) · Stripe MCP docs

PayPal — CAUTION

Payments · rubric-scored; not personally verified · last checked 28 Aug 2026

Official toolkit + MCP across a huge network; does both directions.

Best for: Sending and receiving across a huge global network.

Watch out: In January 2025 New York's Department of Financial Services fined PayPal $2 million over a December 2022 breach that exposed names, Social Security numbers and tax IDs for roughly 35,000 customers; the regulator cited missing multi-factor authentication and rate limiting. Python toolkit support still pending.

Rubric: trust 3/5 · fit 5/5 · ease 4/5 · reliability 4/5 · reach 5/5 · cost 5/5

Sources: NYDFS — $2M penalty, 23 Jan 2025 · PayPal MCP server docs

x402 — CAUTION

Payments · rubric-scored; not personally verified · last checked 28 Aug 2026

Open HTTP micropayment protocol (Coinbase/Visa/Stripe-backed). Crypto only.

Best for: HTTP-native micropayments between agents and services.

Watch out: Crypto/stablecoin settlement only (no fiat yet). A signature-verification bypass in the x402 SDK (GHSA-qr2g-p6q7-w82m, March 2026) has been patched; per the advisory it affected Solana facilitators and did not put user keys, contracts or funds at risk. Separately, independent research demonstrates five practical attacks on the protocol's authorization, binding and replay handling. Real demand still thin.

Rubric: trust 3/5 · fit 4/5 · ease 3/5 · reliability 3/5 · reach 3/5 · cost 5/5

Sources: GitHub advisory GHSA-qr2g-p6q7-w82m — patched · Five Attacks on x402 (arXiv 2605.11781)

Vendor page

Tools with an unresolved, material risk are held off the catalog on a private review bench until a person can verify them. Published tools may still carry disclosed watch-outs — that's transparency, not a gap; the open questions are named right on the card. We also don't rate whole categories of unknown vendors — smaller no-code SMS services (Fleece AI, Whippy, GReminders) and several early agent-payment startups sit there now, unlisted on purpose.
The vetting rubric

Every tool earns its place against six questions.

The same bar, applied to everything. Tools with unresolved risks are held for human review; published tools may still carry disclosed watch-outs, and confirmed risks are flagged as warnings.

01

Trust & Safety

Who makes it? Is it reputable, secure, and clean with your data?

02

Fit

Does it actually do the job the agent needs done?

03

Ease

Can a non-technical person set it up and live with it?

04

Reliability

Will it work every time, dependably, at scale?

05

Reach

Does it deliver where it's needed — the inbox, the phone, the team?

06

Cost

Is it free or affordable to start with?

How the verdict is decided

One of the six decides it. The Trust & Safety score sets the verdict. The other five describe how well a tool fits your situation — not whether it can be trusted. That is why a tool can score well on average and still carry a caution.

Trust 4–5TRUSTEDCleared the bar. Disclosed watch-outs may still apply — read the card.
Trust 3CAUTIONUsable, with caveats you should know before you rely on it.
Trust 1–2WARNINGA known, unresolved problem. Don't use it without reading why.

The six scores are editorial judgement. The step from the trust score to the verdict is not — it is the rule above, applied the same way every time. A verdict is not a human sign-off: a green ✔ means a person at Vettory personally verified that tool, and most entries are not verified and say so. Anything carrying an unresolved material risk is held off this catalog entirely.

For anyone running agents

Point your agent at Vettory.

Ask Vettory what your agent needs, and it returns vetted tools — ranked, scored, and cited, with warnings flagged. No code required to try it.

Try it live · type what your agent needs
Or call it yourself · public demo key
curl "https://vettory-webmcp.misti.workers.dev/api/search?need=send%20email&key=vty_public_demo"
Join the early-access list → Get your own API key and higher limits.