The drafting is the easy half and it is not where quality comes from. What decides whether a message is worth sending is what the agent knew before it started writing, which means the research calls matter more than the model or the prompt.
| Step | The job | Tools tagged | Official MCP | Free tier |
|---|---|---|---|---|
| Research the account | research an account before a call | 17 | 10 | 3 |
| Find a real trigger | detect a funding or news event | 12 | 10 | 3 |
| Scrape the specific page | scrape a web page for facts | 6 | 5 | 2 |
| Draft the message | draft personalized outreach | 51 | 23 | 6 |
| Send it | run an email sequence | 45 | 22 | 3 |
| Read what happened | read outreach performance | 14 | 7 | 1 |
51 entries here are tagged draft personalized outreach, the most tagged job in the entire vocabulary. Drafting is commoditised: every tool in the category will write you a competent paragraph. None of them can invent the fact that makes the paragraph worth reading. That fact comes from a research call, and research is where coverage is thinner: 10 official servers across 17 tagged entries.
12 more are on the linked page. The cut is the display limit, not a ranking.
Personalisation at scale fails in a specific way: the agent finds a fact, the fact is wrong or stale, and the message is now confidently wrong in a way a generic message never would have been. A generic email is ignored. A wrongly personalised one is remembered.
Two guardrails. Make the agent cite the source of every claimed fact in its draft, in a field you can read. And require a human approval on the first send to any account, which costs you almost nothing when the list is good and saves you when it is not.
Which tool writes better copy. There is no tool versus tool verdict anywhere on this site, and 0 of 293 entries have been run by anybody here.
Every number on this page is generated from directory.json at build time and carries the date it was baked: 2026-08-25. Nothing is typed by hand, nothing is rounded, and nothing is estimated. The underlying data is published in full. Where the honest answer is a zero, the zero is printed.