Niches · Workflow automation

Connect an agent to the platform with only the access it needs

Two or three companies' docs show a full fit.

2companies, full fit
12companies, fit with a limit
35companies carry the tag
11share of the category's work
7first-hand accounts

What counts as done

The MCP or API connection authorises without a browser step after the first, the tool list loads in the client, and a test shows tools outside the grant are absent and a write outside it is refused.

How agents ask for it: connect Claude to my n8n, add the Activepieces MCP server, give this agent run-only access to two workflows

Does the job 2

The docs show it meets the test above, and nothing we read says otherwise.

ListingWhy
Make API + MCP B
Celonis, Inc.
OAuth at mcp.make.com/v2 with scopes chosen at connect, and scopes decide which tools load; an API key in a header can be limited to an organisation, team or named scenarios (now MCP toolboxes). source
Windmill API + MCP D
Windmill Labs, Inc.
OAuth sign-in picks workspaces and an access level (Build and run, Read and run, Read only, or named scripts and flows as tools, which give the client nothing else), tokens go in a header, and calls outside the user's permissions are refused. source

Does the job, with a limit 13

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
Activepieces API + MCP C
Activepieces Inc.
OAuth MCP with tool groups toggled per project, but one tool list per project for every client (acc-16), empty protected-resource metadata on self-hosted (acc-17) and a Cloudflare block for ChatGPT (acc-15). source
Microsoft Copilot Studio MCP D
Microsoft Corporation
Since 1.0.4 sign-in persists in an OS-protected cache across chats, but it needs a local desktop browser (no SSH or headless), and access is the user's full role: no grant narrows the tool list. source
Gumloop C
AgentHub Inc. (doing business as Gumloop)
Remote MCP at mcp.gumloop.com with 87 tools; a team-scoped connection drops 33 personal-key tools. The only API OAuth scope is gumloop_api, so a grant can't be limited to running named agents. source
Kestra C
Kestra Technologies SAS
Each MCP server lists only flows pointed at it, which narrows tools, but open-source servers take Basic auth only; API_TOKEN and OAuth (needed for Claude web) and RBAC are Enterprise or Cloud. source
Kognitos C
Kognitos, Inc.
OAuth 2.1 or a header API key; keys are scoped to workspaces with All, Read only or Restricted per-resource permissions. The docs don't show the MCP tool list shrinking under a narrow key, and OAuth needs callback port 3777. source
n8n API + MCP D
n8n GmbH
OAuth or a personal token, and API keys take scopes from 2.14.0, but MCP access isn't scoped per client (every client sees every MCP-enabled workflow), Gemini rejected the tool schemas (acc-38), and Chrome flagged the consent page (acc-39). source
Node-RED B
OpenJS Foundation
No MCP server. The Admin API issues tokens by password grant with scope * or read, lasting seven days, so a read-only grant exists, but nothing narrower and no run-only grant. source
Paragon ActionKit + MCP D
Forge Technology, Inc. (d/b/a Paragon)
A per-user JWT narrowed by the urn:useparagon:connect:permissions claim limits credentials and call types (ActionKit, Proxy, workflow triggers), but the listing records that the self-hosted MCP server runs in development mode unless NODE_ENV=production is set. source
Pipedream API + MCP BB
Pipedream, LLC
Client-credentials OAuth tokens take a scope parameter (default *), and x-pd-app-slug loads one app's tools, but a granular-scope token got 403 on reloadProps where connect:* worked (acc-27). source
Microsoft Power Automate C
Microsoft Corporation
No MCP server on the pages read; Dataverse Web API access runs under the caller's security roles, so a narrow grant is a role, and the agent's tool list is whatever its client builds. source
Prismatic C
Prismatic Software Inc.
Instance- and integration-scoped MCP URLs narrow the tool list, but API tokens carry the creating user's whole role (a guest user is the narrow option), and the dev server needs prism login in a browser once. source
Tray.ai API + MCP C
Tray.ai, Inc.
OAuth2 binds the session to one workspace, and Agent Gateway serves only the tools a builder exposes, with an access list. Headless MCP acts with the user's full permissions, deletes included, and API-token clients need an npm bridge. source
Workato API + MCP C
Workato, Inc.
OAuth uses the signed-in user's permissions in one workspace and environment, but an API token runs as the workspace owner whatever the client role's other privileges, and OAuth needs a browser per workspace. source

Doesn't do this job 3

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

ListingWhy
Bind AI F
1Fort Inc.
No MCP server, OAuth client registration or scoped agent token is documented; access is a user JWT or an Api-Key header with no scope list. source
Cooper (Mandl) F
Cooper AI Tech, Inc.
No MCP server, API key or OAuth client for an outside agent. source
Motive Agents F
Motive Technologies, Inc.
No MCP or API surface for Agents themselves; Motive's MCP server covers fleet data, which is a different listing. source

What stops agents finishing

  • OAuth metadata endpoint empty or blocked by a bot check
  • grant cannot be narrower than the server's full scope list
  • one tool list per project for every client
  • browser sign-in after every restart (Copilot Studio)
  • consent page flagged by Safe Browsing

Around it

Before: chat.user-asks. After: workflow-automation.build-workflow, workflow-automation.run-workflow-as-tool.

Where this comes from

Weight 11 of 100 in Workflow automation: 7 of 48 accounts (n8n, Activepieces, Pipedream); MCP OAuth failures in GitHub issues and the n8n forum

The same as data: /api/v1/use-cases/workflow-automation.connect-agent.json, or the MCP tool get_use_case with id workflow-automation.connect-agent. Other jobs in this category: /api/v1/niches?category=workflow-automation.

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.