Merge Agent Handler alternative for AI agents acting in users' accounts

Merge Agent Handler is an MCP server that connects your agent to 190+ tools, stores each user's credentials, scans every call for sensitive data, and logs who called what. Anlyon connects a shorter list and controls each write per user: a limit you set, counted before the call, the user's yes in your app bound to the exact request, and a graded receipt read back from the provider.

Merge has an agent product with its own docs, so this page compares against it directly. Facts about Merge come from Merge's own docs and site, checked on 2026-11-02.

What Merge Agent Handler does

Merge calls Agent Handler "a platform for securely connecting AI agents to enterprise tools at scale" (product page, as of 2026-11-02). The docs cite "190+ MCP-ready third-party connectors" (overview, as of 2026-11-02).

Per-user auth. "There is one Registered User per end user." The user authenticates a connector through Link, an embedded UI, and "Agent Handler stores the credential against the Registered User; every subsequent call uses it transparently." "OAuth refresh happens here if the access token is near expiry" (how it works, as of 2026-11-02). Merge makes the third-party call.

Binding the user. Your backend mints a short-lived user-scoped Access Key that names one Registered User and the Tool Packs it can reach. Merge checks registered_user_ids "on every tool call, against the key rather than against the URL", and editing the user in the MCP URL fails with a 404 (scoping access per user, as of 2026-11-02). This is a solid design: the agent's session cannot reach another user.

Security Gateway. Entity and regex rules "take allow, redact, or block on tool arguments" before they leave Merge, and AI Guardrails "take block or log" and also screen responses (how it works, as of 2026-11-02).

Limits. MCP tool calls are "Limited per organization and per Registered User" (API overview, as of 2026-11-02). Merge's FAQ answers "Can I set per-user rate limits?" with "Not as a self-serve setting", and says the built-in per-user rate is "a per-minute rate rather than a monthly cap, so it does not stop one user from spending the month's credits" (FAQ, as of 2026-11-02).

Approvals. In the Agent Handler for Employees setup, an employee can request a tool and an admin can "Approve permanently or with an expiry" (managing tool access, as of 2026-11-02). That grants access to a tool, not a single call. We did not find an approval of a single tool call by the end user in Merge's Agent Handler docs (docs index) as of 2026-11-02.

Audit, deletion, self-host, pricing. Tool call logs and an audit trail are kept for a retention window, "The default is 90 days" (audit trail, as of 2026-11-02). DELETE /api/v1/registered-users/{id} "removes the profile and credentials", with tool-call and audit records kept for the retention window, then purged (registered users, as of 2026-11-02). Standard plans run on Merge's hosted infrastructure, and Enterprise contracts can include single-tenant instances and VPC peering (FAQ); the pricing page lists "Custom deployments, including on-prem" on Enterprise. Free is $0 with 2,000 monthly credits, Pro is $1000 a month with 25,000 credits, and a credit is one tool call (pricing, as of 2026-11-02).

What Anlyon does

Anlyon starts in the same place: each user connects their own account, Anlyon holds the token, and your server names the user. Then it controls the write itself.

const user = anlyon.forUser(session.user.id);
const r = await user.actions.invoke('hubspot-update-deal-stage', {
  dealId: '123',
  dealstage: 'contractsent',
});
// Later, read the graded receipt for that write:
// await user.actions.effect(effectId)  ->  { id, outcome, grade, impact }

A limit you create with scope: 'end_user' is counted and reserved before the call, in money, record changes or a unit you declare, such as emails. A policy with decidedBy: 'end_user' sends a risky step to that user to decide in your app, and the approval is bound to a SHA-256 digest of the exact request, rechecked before dispatch. Each governed write returns a receipt graded confirmed (a read-back of the provider's own record matched), acknowledged, unknown, failed, refused or denied. Anlyon never resends an unknown write.

Side by side

Anlyon's limits, stated once: the list of integrations is shorter, and Gmail, Google Calendar and Salesforce are not live yet. End users are supported over the HTTP API and the TypeScript and Python SDKs, not yet over MCP or the CLI. Anlyon is a small company.

When to choose Merge Agent Handler

• Your agent speaks MCP and you want one endpoint for 190+ connectors.
• You need DLP scanning, redaction and guardrails on tool inputs and outputs.
• You also use Merge's Unified API, which shares the same end-user credential store.
• You want employees' AI clients provisioned through SCIM with tool access requests.
• You need single-tenant, regional or on-prem deployment under an enterprise contract.

When to choose Anlyon

• Your agent writes in your users' accounts and you want to set each user's ceiling yourself, by day or month, in money or record changes.
• A risky step should wait for that user's yes in your app, and the request that runs must be the one they approved.
• You need to know whether each write landed, graded against the provider's own record.
• You want a connect link that a forwarded click cannot attach to the wrong user.
• You want to erase a user in one call. See acting for your users, per-user limits and user approvals in your app.

Using both

Keep Agent Handler for its catalogue and its scanning, and route the writes you want limited, approved and graded per user through Anlyon. Your server holds both keys and uses the same user id with each. Each user connects to each product separately, and a credential in one is not available to the other. For a wider view, see agent auth platforms compared and Arcade, Composio and Anlyon.

Sources

All checked on 2026-11-02.

• Agent Handler product page: https://www.merge.dev/agent-handler
• Overview: https://docs.merge.dev/merge-agent-handler/overview.md
• How it works: https://docs.merge.dev/merge-agent-handler/how-it-works.md
• Scoping access per user: https://docs.merge.dev/merge-agent-handler/secure/scoping-access-per-user.md
• API overview and rate limits: https://docs.merge.dev/merge-agent-handler/agent-handler.md
• FAQ: https://docs.merge.dev/merge-agent-handler/resources/faq.md
• Managing tool access: https://docs.merge.dev/merge-agent-handler/administer/managing-tool-access.md
• Tool call logs: https://docs.merge.dev/merge-agent-handler/observe/tool-call-logs.md
• Audit trail: https://docs.merge.dev/merge-agent-handler/observe/audit-trail.md
• Registered users: https://docs.merge.dev/merge-agent-handler/build/users/registered-users.md
• Docs index: https://docs.merge.dev/merge-agent-handler/llms.txt
• Pricing: https://www.merge.dev/pricing/agent-handler
• Anlyon docs: https://docs.anlyon.com/guides/act-for-your-users