Niches · Agent auth & delegated access

Connect an agent to a user's account in a third-party app

One company's docs show a full fit.

1companies, full fit
10companies, fit with a limit
16companies carry the tag
25share of the category's work
12first-hand accounts

What counts as done

After one consent, the agent gets a valid token or a proxied call for that user and provider, bound to the right user ID, with the granted scopes returned, and the first API call succeeds.

Measured, for our own tests: the first API call succeeds within 2 calls of the consent completing. No docs we read state these figures, so they don't count against a listing here.

How agents ask for it: connect my Gmail so the agent can read it, let the agent post to our Slack as me, give each customer's agent access to their HubSpot, link my GitHub account

Does the job 1

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

ListingWhy
WorkOS Pipes and Agents B
WorkOS, Inc.
After the user connects through the authorize URL or Pipes widget, POST /data-integrations/{slug}/token with user_id returns active, a refreshed access token, its scopes and missing_scopes. Custom OAuth credentials and custom providers are supported. source

Does the job, with a limit 10

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
Aembit BB
Aembit, Inc.
MCP User-Based Access Token stores and refreshes each user's tokens and proxies calls through the MCP Identity Gateway, keyed to the IdP-verified user. Covers MCP servers only, no granted-scope list is returned, and the managed gateway is set up through Aembit. source
Amazon Bedrock AgentCore Identity A
Amazon Web Services, Inc.
GetResourceOauth2Token returns the user's token for built-in or custom OAuth providers after consent. The ForUserId path takes an unverified user string, granted scopes are not returned, and AWS says a returned token may already be revoked. source
Arcade.dev B
Arcade AI, Inc.
tools.authorize returns status, url, scopes and user_id, then the tool runs with only user_id. Arcade's default OAuth apps accept only members of your Arcade project, so real users need your own OAuth app and a custom user verifier first. source
Auth0 for AI Agents (Token Vault) BB
Okta, Inc.
After the Connected Accounts flow, a Token Vault exchange returns the provider access token with its scopes and expiry for that Auth0 user. Free allows 2 Token Vault connections and Essentials 3 without the add-on, and Auth0 must be the identity provider. source
Descope Agentic Identity Hub A
Descope, Inc.
Connections vault each user's provider tokens and the agent fetches them with that user's own access token under policy, or gets ConnectionAuthorizationRequired with a connect URL. The provider token is handed to the agent, and the Agent Auth SDK is still not on PyPI or npm. source
Keycard D
Keycard Labs, Inc.
A tool exchanges the user's Keycard token for a short-lived upstream token on every call, and grants record the approved scopes per provider. A failed exchange lands on AccessContext rather than raising, so code that skips has_errors() calls on with nothing; sign-up is by request during Early Access. source
Microsoft Entra Agent ID BB
Microsoft Corporation
An agent gets user-delegated tokens through the OBO flow for Microsoft Graph, Azure and APIs registered in Entra. There is no store of third-party grants such as Gmail or Slack, and agent identities cannot run interactive consent themselves (AADSTS82014). source
Nango B
Nango Inc
A connect session links the user's account, and an agent session exposes that tenant's connections as an MCP server through the proxy, so the agent never sees raw credentials; a first-hand account reports a year of stable use. Free stops at 10 connections. source
Scalekit AgentKit BB
ScaleKit, Inc.
An authorization link activates the user's connected account, then tools run with identifier plus connection name across 672 connectors. User verification starts at None, where anyone holding the link activates the account, so a custom verifier is needed before real users. source
Vercel Connect BB
Vercel Inc.
getToken with a user subject returns a short-lived provider token, and UserAuthorizationRequiredError leads to startAuthorization for consent; the app never sees refresh tokens. The user ID comes from the calling deployment, and Hobby includes only 500 token requests a month. source

Doesn't do this job 3

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

ListingWhy
Clerk MCP server B
Clerk, Inc.
The server signs the agent in to the developer's own Clerk workspace only; it holds no third-party provider grants for end users. source
Cloudflare Workers OAuth Provider A
Cloudflare, Inc.
It is an authorisation and resource server for the developer's own API or MCP server; upstream providers appear only as sign-in options, with no store of users' third-party grants for an agent to call. source
Stytch Connected Apps B
Twilio Inc.
Connected Apps makes your project an authorisation server for clients calling your app; it has no outbound vault for users' third-party tokens. source

What stops agents finishing

  • provider not in the catalogue and no generic OAuth connector
  • free tier caps connections (Auth0 Token Vault: 2 on Free)
  • default OAuth app only accepts the vendor project's own members
  • consent needs a person in a browser at the moment of the request

Around it

Before: chat.user-asks, crm.find-account. After: email.read-inbox, messaging.post-message, crm.update-record.

Where this comes from

Weight 25 of 100 in Agent auth & delegated access: 12 of 33 accounts (MCP production thread, Nango, Auth0, Arcade, AgentCore, tenant-routing comments); the most described job, scaled from overlapping counts

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

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.