Niches · Agent tool access

Do one thing in an app the person has connected

No company's docs show a full fit yet. The listings below do the job with a limit, or not at all.

0companies, full fit
8companies, fit with a limit
36companies carry the tag
22share of the category's work
8first-hand accounts

What counts as done

The action runs, every field the person asked for is set, and the result (record id or link) is returned; a failure names the app's own error.

Measured, for our own tests: it runs in one or two tool calls. No docs we read state these figures, so they don't count against a listing here.

How agents ask for it: add this to my Airtable, create a project in Kantata, update the sheet with these numbers, send this to the team in Slack

Does the job, with a limit 8

It does the job, but something in the docs or a user's account will get in an agent's way. The limit is named in each row.

ListingWhy
AgentWeb D
Agent Web Labs, Inc.
execute runs one pre-mapped website action and returns a verified state and receipt, with structured errors. Fields are limited to what AgentWeb's map declares, there is no raw request fallback, and coverage is consumer sites rather than app APIs. source
Composio (API + MCP) A
Sampark Inc (d/b/a Composio)
Session tools execute app actions for a userId, and Proxy Execute sends a raw authenticated request when a tool lacks a field. acc-23 (still open) shows a provider rejection surfacing only as invalid_request, and acc-22 shows integer ids rounded until an October fix. source
MCP Gateway BB
No legal entity found. COMMERCIAL.md, CLA.md and AUTHORSHIP.md name Mikko Parkkola, Finland, as licensor and author
gateway_invoke routes a call to a backend MCP server or REST capability the operator configured, so fields are whatever the backend exposes. Arguments aren't checked against the backend schema, and each task takes an extra search turn. source
Merge Agent Handler BB
Merge API, Inc.
Tool Packs expose connector tools run with the Registered User's credentials, and logs keep arguments and results. Field coverage per connector and whether the app's own error reaches the agent weren't documented in the pages read. source
One B
One Systems, Inc.
execute_one_action runs an action on a connected app and failures come back as readable tool errors. The usual flow is three tools, skippable to two when the grant lists actions, and acc-27 shows Sheets reads failing output-schema validation until a 20 October fix. source
StackOne B
StackOne Technologies Limited
Actions across 540+ apps run against a linked account named in x-account-id, with request logs and a request tester for debugging failed calls. Field coverage per connector and whether the provider's own error text reaches the agent weren't checked. source
Unified.to MCP Server B
Unified API Inc.
A connection's MCP server turns the integration's supported objects into tools on the unified API, so fields are the unified model's, not every native field. Whether the app's own error reaches the agent wasn't documented in the pages read. source
Zapier MCP (agent actions) C
Zapier Inc.
execute_zapier_read_action and execute_zapier_write_action run actions across 9,000+ apps after discover and enable. acc-13 shows Kantata fields the action didn't expose, acc-12 a halted error hiding a bad argument, and acc-16 both execute tools failing input validation. source

Doesn't do this job 3

Listings in the category that do something else, with the reason.

ListingWhy
letme BB ours
Long Lunch Labs Limited (company number 17523112, registered in England and Wales, incorporated 15 October 2026), trading as Anchor Terminal and letme
letme picks a tool and says how to call it; it forwards nothing, holds no app connections, and a POST to any path but / and /mcp returns 405. source
Monid C
Monid, Inc.
Monid's own docs send app writes to Composio or Zapier; it runs paid data endpoints, not actions in a person's connected apps. source
Smithery F
Clavia, Inc. (doing business as Smithery)
The home page says Smithery's MCP Proxy and managed identity shut down on 31 October 2026; there is nothing left to run actions through. source

1 listings in the category were added after this reading and aren't marked yet.

What stops agents finishing

  • action omits a field the task needs
  • tool schema rejected by the model API
  • misleading error hides a bad argument

Around it

Before: aggregator.connect-end-user, aggregator.find-tool. After: notifications.push, spreadsheets.read-back.

Where this comes from

Weight 22 of 100 in Agent tool access: 8 of 32 accounts (Zapier community threads, Composio, One, Smithery-hosted toolkit), tied highest

The same as data: /api/v1/use-cases/aggregator.run-app-action.json, or the MCP tool get_use_case with id aggregator.run-app-action. Other jobs in this category: /api/v1/niches?category=aggregator.

Draft, read 2026-11-15 from public docs and first-hand accounts, not yet reviewed by a person. A mark says what the docs show, not what a run measured.

For companies

Do agents find, use and choose your tools?

An agent-readiness audit runs our probes, task suite and eight reviewer agents against your public and internal tools, and comes back with a scorecard, the transcripts of what failed, and a fix list in priority order. From $2,500, re-run included. We never take payment to move a rank. We do help companies earn one.