Last verified: 2026-07-27 · Disclosure
Historically this dimension meant "how many integrations does the marketplace list," which was always a vanity metric. It now means something concrete: can an agent do everything a person can do in this tool?
The single most important property, and the one nobody advertises. An API covering 60% of what the interface can do caps everything you or your agents can automate — permanently, regardless of how good the model is.
Test it directly: pick the three actions your workflow depends on and check whether each has an endpoint. Vendors describe their API as "comprehensive" without exception.
The difference between an API that supports an integration and one that supports an agent. An agent doing real work over a large workspace makes far more calls than a webhook-driven integration, and limits designed for the latter make the former unusable.
Notion's roughly-three-requests-per-second is the clearest example here: fine for a Zapier connection, painful for bulk migration or an agent reasoning across a large workspace.
The emerging standard for exposing a tool to agents. A first-party MCP server means Claude or any MCP client can operate the tool directly, without you writing glue code. In two years this will be a checkbox; today it's a real differentiator and a decent proxy for whether a vendor is thinking about agents at all.
Caveat worth checking: an MCP server wrapping a partial API inherits the gap. Coverage matters more than existence.
Reliability, delivery guarantees, retries, and event coverage. Less glamorous than the API and more likely to be the thing that breaks your automation at 3am.
The strongest form of extensibility, and the strongest mitigation for vendor risk. Among tools here: Jira Data Center (self-hosted, not open source), Macro (AGPLv3, self-hostable), Obsidian (local by default). Everyone else, you're a customer of a hosted service.
| Tool | API quality | Rate limits | MCP | Self-host | Ecosystem | Score |
|---|---|---|---|---|---|---|
| Slack | Excellent, mature | Generous, tiered | Community | No | Largest in this repo | 5 |
| Jira | Complete, older | Reasonable | Via Atlassian tooling | Data Center | Thousands of apps | 5 |
| Airtable | Excellent REST + scripting | Moderate | Community | No | Large | 5 |
| Linear | Excellent GraphQL, near-parity | Good | Official | No | Small, curated | 4 |
| Attio | Clean, well-documented | Good | Yes | No | Growing | 4 |
| Coda | Good, plus Packs SDK | Moderate | Limited | No | Modest | 4 |
| Macro | Full, MCP at UI parity | Good | Yes, first-class | AGPLv3 | Small | 4 |
| Asana | Solid REST | Reasonable | Community | No | Large | 4 |
| Monday | Good GraphQL | Metered actions | Community | No | Large | 4 |
| HubSpot | Comprehensive | Tiered | Community | No | Very large + consultants | 4 |
| Obsidian | Local plugin API | N/A | N/A | Local files | Thousands of plugins | 4 |
| Notion | Adequate | ~3 req/s — restrictive | Official | No | Large | 3 |
| ClickUp | Broad but inconsistent | Moderate | Community | No | Large | 3 |
| Height | Exists | Unknown | Limited | No | Small | 3 |
| Superhuman | Minimal | — | No | No | None | 2 |
| Basecamp | Basic | Reasonable | No | No | Small | 2 |
Slack's 5 is about the ecosystem, not the API. The API is good; the reason it scores 5 is that the app directory is genuinely unmatched and Workflow Builder lets non-developers automate without engineering. If your operations depend on a specific Slack app, that fact alone can decide a chat migration.
Jira's 5 is about the marketplace and Data Center. Thousands of apps, many of them mature businesses, plus the only real self-hosting option among established players. For organizations that cannot use cloud SaaS this narrows the field to nearly nothing.
Linear's 4 is the highest-quality API here per unit of surface area. Well-designed GraphQL, near-complete parity, and an official MCP server. It scores 4 rather than 5 because the ecosystem is deliberately small — that's a product decision, not a defect, and it still means fewer prebuilt things exist.
Notion's 3 is entirely the rate limit. The API is otherwise fine and the ecosystem is large. Three requests per second is the thing that will frustrate you, and it's the most common complaint from anyone building on Notion seriously.
Monday's 4 has an asterisk: automation and integration actions are metered per tier. The API is good and your usage of it has a monthly ceiling that varies with what you're paying. Check your tier's limit against your intended usage before building anything.
Macro's 4: full MCP surface built at parity with the UI rather than as an afterthought, and AGPLv3 with a self-host path — the strongest position here on vendor risk. It scores 4 not 5 because the ecosystem is genuinely small and the prebuilt integrations Slack and HubSpot users take for granted mostly don't exist. We build this; the ecosystem gap is real and not something we'd argue away.
If you intend to point agents at your workspace, four things to verify before committing — none of which appear on a pricing page:
- Does an official MCP server exist, and what fraction of the UI does it cover? Existence isn't coverage.
- What are the rate limits, and what happens at the ceiling? Hard block, throttle, or overage charge.
- Can an agent authenticate as a user with that user's permissions, or only as a workspace-level bot? The second is a security problem waiting to be discovered, because a bot token that can read everything will eventually read something it shouldn't.
- Is there an audit trail of agent actions? When an agent changes 200 records, you need to know which ones and be able to reverse it.
Question 3 is the one most vendors currently fail, and it's the one that will matter most as agents take more actions. Ask it explicitly.