MoEngage has launched Merlin AI Custom Agents, positioning the feature set around marketer-defined guardrails, step-by-step activity visibility, and an open Model Context Protocol (MCP) server for connecting with external AI systems.
The release targets a common adoption blocker for agentic marketing workflows: teams want automation, but they also need auditability, controllable permissions, and the ability to run in assisted mode rather than full autopilot.
Table of contents
Jump to each section:
- What MoEngage launched and how it works
- Why “visibility plus guardrails” is becoming table stakes for agentic martech
- Open MCP architecture and what it changes for enterprise stacks
- Competitive context in customer engagement platforms
- Practical takeaways for lifecycle and CRM teams
What MoEngage launched and how it works
Merlin AI Custom Agents lets lifecycle and CRM teams design agents that run continuously using MoEngage’s own data and tools, within rules the marketer sets. The product framing emphasizes “show your work” activity logs, including the data pulled, the decisions made, channels touched, and the content sent.
Alongside custom agents, MoEngage also added three Merlin agents:
- In-app template generator to produce in-app templates with responsive code and interaction logic based on a description.
- Flows assist to draft multi-stage journeys with triggers, paths, and channel steps for marketer review.
- Campaign insights agent to answer questions about campaign performance in plain language and surface recommended changes.
A key workflow detail is that teams can choose operating modes: full autonomy or human review before execution. That matters because many orgs want to start with “copilot” patterns, then expand autonomy as governance matures.
ContentGripWhy “visibility plus guardrails” is becoming table stakes for agentic martech
Marketing agents only scale if they can be trusted inside production systems that handle brand risk, regulatory constraints, and high message volumes across email, push, in-app, and SMS. In practice, “trust” often comes down to operational controls: who can let an agent send, what budgets or frequency caps it must obey, which audiences it can touch, and whether every action is reviewable after the fact.
MoEngage’s emphasis on full visibility and marketer-defined limits reflects a broader shift in AI marketing automation: buyers are increasingly evaluating AI not just on output quality, but on governance features that make AI safe to run continuously. For teams managing large-scale lifecycle programs, an agent that cannot be audited is hard to deploy beyond small experiments.
Open MCP architecture and what it changes for enterprise stacks
MoEngage is also introducing an MCP server and “agent-callable” APIs so external AI systems such as Claude and ChatGPT can access MoEngage context and tools, and so external agents can coordinate with Merlin agents.
Strategically, this is a bet on interoperability rather than a closed “all-in-one” AI layer. For enterprises that are standardizing on certain AI assistants or orchestration layers, an MCP-style connector can reduce integration friction and help MoEngage fit into an existing stack, instead of requiring teams to rebuild workflows around a single vendor’s interface.
The practical implication for marketers is that “AI in martech” is starting to look more like a set of composable services: analytics context, orchestration actions, creative generation, QA checks, and reporting can be distributed across systems, as long as there is a secure protocol for passing context and invoking actions.
Competitive context in customer engagement platforms
MoEngage competes in a crowded customer engagement and marketing automation category where vendors differentiate on omnichannel orchestration, personalization depth, analytics, and workflow automation. Competitive platforms in this space include Braze, CleverTap, Insider, and Airship.
Merlin AI Custom Agents pushes differentiation toward agent operations: controllable autonomy, transparency of decisioning, and the ability to integrate with external AI tooling via MCP. If those capabilities work reliably at scale, it can appeal to teams that want to automate repetitive lifecycle tasks (QA, journey drafting, performance reporting) without turning campaign execution into an opaque “black box.”
At the same time, the bar is rising across the category. As AI-assisted journey building and natural-language insights become more common, defensibility will likely depend on how well agents connect to first-party data, respect governance constraints, and prove measurable performance improvements without increasing brand or compliance risk.
Practical takeaways for lifecycle and CRM teams
- Start with bounded agent scopes. Use custom agents first for QA checks, report generation, or draft journeys, where errors are easier to catch before sending.
- Define guardrails like you would for humans. Specify audience eligibility, channel restrictions, frequency caps, and budget limits so the agent’s operating envelope is explicit.
- Treat activity logs as a core requirement. If an agent cannot explain which data it used and what actions it took, troubleshooting and stakeholder buy-in get harder.
- Plan for cross-system workflows. If your org already uses external AI assistants, an MCP connector can matter as much as in-product agents because it reduces duplication of logic across tools.
- Benchmark against existing platforms. Evaluate whether agent features reduce campaign cycle time or analyst workload versus what you can already do in Braze-like or CleverTap-like workflows.
Leave a Reply