Merge Agent Handler alternative. StackOne vs Merge Agent Handler
Merge Agent Handler hands developers an MCP URL per Tool Pack and user. StackOne connects employees' AI and your agents to 640+ apps through one Tool Gateway MCP URL. Each person signs in once with OAuth 2.1 through the StackOne login or your SSO, and StackOne stores and refreshes their tokens. Defender scans every tool response for prompt injection, and Team and Enterprise plans add field-level policies and PII redaction on each call. Self-host it from a supplied Helm chart and Terraform when the data cannot leave your cloud.
Compared on . Merge rows cite Merge's public docs.
Merge Unified vs Agent Handler. Merge Unified and Agent Handler are separate products. This page compares Agent Handler.
A normalized data API
Merge Unified maps each category of system, such as HRIS or ATS, onto one common schema, so application code reads the same fields from many providers through one integration. For why a fixed schema is a poor fit for tool calls, see 7 unified API limitations for AI agents.
Tool calling for agents
Agent Handler gives an agent MCP tools across Merge's connectors, with per-user credentials, Tool Packs that scope what each agent surface can reach, and a Security Gateway that checks tool calls. End users connect their accounts through Merge Link, as in Merge Unified.
Side by side. Merge Agent Handler and StackOne, feature by feature
The table compares how each product connects apps, secures each tool call, keeps the model's context small and deploys. Merge entries come from Merge's public docs and pricing page, listed in the sources below.
| Capability | Merge Agent Handler | StackOne Start free |
|---|---|---|
| Connect Which apps employees' AI and your agents reach, and how each person signs in. | ||
| One MCP URL for every AI client | Employees sign in with SSO after SCIM provisions them from your identity provider, and tools are assigned per IdP group. Developers get a separate URL for each Tool Pack and Registered User, sent with an API key. | One Tool Gateway URL works in Claude, Microsoft 365 Copilot, ChatGPT, Gemini Enterprise, Glean and Cursor. Each person signs in once with OAuth 2.1. |
| Pre-built connectors | Hundreds of MCP-ready connectors (Merge docs, Nov 2026) | 640+ connectors, 46,000+ actions |
| Tool list published per connector | Included | Included |
| Per-user authentication | Registered Users connect each app through Link or Magic Link | Each person signs in with OAuth 2.1 and PKCE through the StackOne login or your SSO and picks which linked accounts a client can use. StackOne stores and refreshes the tokens, so the assistant never sees a credential. |
| Use your own OAuth apps | Included | Included |
| Custom connectors | Register a remote MCP server as a connector. Custom connector builds are on Enterprise. | Any remote MCP server can become a governed connector, and AI Connector Builder customizes actions down to the field (Team and Enterprise). |
| Secure What each tool call is allowed to do, and what reaches the model. | ||
| Tool access | Tool Packs set the tools each agent surface can reach | Allow or deny tools by connector category, connector, account or tool name, for everyone, users or groups (Team and Enterprise) |
| Field-level read and write rules | Not documented | Read a field (masked in responses) and Write a field (Team and Enterprise) |
| Rules on argument values | Custom regex rules on tool arguments | Value rules on the arguments a tool receives (Team and Enterprise) |
| PII detection | Entity rules (identity, government and financial IDs, payment, health, network) on tool arguments: allow, redact or block | StackOne's own PII model finds email, phone, SSN and card numbers inside any field, free text included, then redacts or blocks them on inputs and responses (Team and Enterprise) |
| Plain-language rules | AI Guardrails on requests and responses: block or log the call | Semantic rules: refuse the call on inputs, mask the matching field on responses (Team and Enterprise) |
| Prompt injection scan on tool responses | Not documented | Defender scans every tool response: monitor, sanitize or block |
| Try a rule before enforcing it | Allow and Log actions, plus a rule tester | Test a policy against chosen members without making a call, then Monitored mode records what it would redact or block before Enforced mode acts (Team and Enterprise) |
| Logs and SIEM | Tool Call Logs and an audit trail, with SIEM streaming on Enterprise | Action logs on every plan, the Logs API on OEM plans and Enterprise, and SIEM export on Enterprise |
| Optimize How much of the model's context the tools take as the catalog grows. | ||
| Tools in the model's context | tools/list returns the Tool Pack's tools plus a search_tools meta-tool | Two tools, search and execute, so the context stays the same size as the catalog grows |
| Platform Where it runs, what it is certified for, and how to start. | ||
| Deployment | Managed cloud, or on-prem and mVPC on Enterprise | Managed cloud in the EU or US, a dedicated account, or self-hosted on any Kubernetes cluster from a supplied Helm chart and Terraform (Enterprise) |
| Agent protocols | MCP | MCP and A2A |
| Compliance | SOC 2 Type II, ISO 27001, HIPAA, GDPR, CCPA | SOC 2 Type II, HIPAA, GDPR, CCPA |
| Free plan | Free plan with 2,000 credits a month | Starter: 1,000 credits per seat a month, no credit card. OEM Core, for agents in your product: no platform fee |
Merge sources, checked Nov 3, 2026
Cells read "Not documented" where Merge's docs do not cover a capability.
- Connector count, compliance, audit trail: Agent Handler overview (docs.merge.dev/merge-agent-handler/overview)
- Developer MCP URL, API key auth, tools in context: MCP integration (docs.merge.dev/merge-agent-handler/build/connecting-agents/mcp-integration)
- Employee MCP URL, SSO, SCIM and IdP groups: Agent Handler for Employees (docs.merge.dev/merge-agent-handler/setup/agent-handler-for-employees)
- Per-user auth, OAuth apps, Tool Packs, tool lists, logs: How it works (docs.merge.dev/merge-agent-handler/how-it-works)
- Remote MCP servers as connectors: Custom MCP servers (docs.merge.dev/merge-agent-handler/build/connecting-agents/custom-mcp-servers)
- Entity rules, regex rules, rule tester: Security Gateway (docs.merge.dev/merge-agent-handler/secure/security-gateway)
- Plain-language rules: AI Guardrails (docs.merge.dev/merge-agent-handler/secure/ai-guardrails)
- Custom connector builds, SIEM, deployment, free plan: Agent Handler pricing (merge.dev/pricing/agent-handler)
StackOne rows are documented in Tool Gateway, OAuth client setup, Tool Policies, PII redaction, Defender, your own OAuth apps, MCP connectors, A2A, observability, deployment options and pricing.
The decision. Where each one is the better fit
- You already run Merge Unified and want agents on the same Merge Link flow.
- ISO 27001 is on your vendor checklist.
- Employees use more than one AI client. One Tool Gateway URL serves Claude, Microsoft 365 Copilot, ChatGPT, Gemini Enterprise, Glean and Cursor, and each person signs in once through the StackOne login or your SSO.
- Your security team wants tool policies that reach individual fields, to mask what an agent reads and block what it writes.
- You want every tool response scanned for prompt injection before the model reads it, with Defender.
- Your agents need 46,000+ actions across 640+ connectors without filling the context, because they reach them through two tools.
- PII has to be caught inside free text such as ticket notes. StackOne's own PII model redacts it in tool inputs and responses, and you can watch what it would redact in Monitored mode before enforcing.
- You want to self-host from packaged infrastructure: StackOne supplies a Helm chart and Terraform for AWS, Google Cloud, Azure or any Kubernetes cluster, plus releases and connector updates (deployment options).
- Your agents talk A2A as well as MCP.