What Melaya learned from its first 754 MCP tool calls

Imagine asking your assistant to open the supplier portal you have selected and collect last month's invoices into a table. Not explain how to do it. Work through the page and bring back the information.

That is the kind of workflow Melaya ↗ is built for. Its remote MCP server gives compatible AI assistants access to browser actions, Android apps, connected business tools and agent workflows, with permissions chosen by the user.

For an operations team working across portals, or a founder juggling browser and phone tasks, the appeal is simple: less carrying information between the assistant and the systems where work happens. The invoice task is an illustrative example of Melaya's capabilities, not a customer result measured in this story.

Melaya makes that work possible. Melaya uses Flowlines for MCP analytics and observability, tracking tool adoption, user intent and release changes. Here is what the partnership revealed in the first 48 hours of MCP telemetry.

Work in the browser and apps you already use

An assistant is more useful when it can reach the place where the task lives. Melaya extends that reach beyond connected APIs to selected browser tabs and allowed apps on a paired Android device.

In the browser, that could mean collecting records from a supplier portal. On Android, it could mean reviewing unread messages and summarizing which ones need attention, without sending a reply. These are different surfaces, accessible through the same Melaya MCP connection.

You keep your preferred compatible assistant rather than moving the whole workflow into a new chat interface. Melaya supplies the tools and the execution environment. Browser control uses its extension; phone control is for Android, not iOS.

Turn repeat requests into repeatable workflows

Some jobs should not start from a blank conversation every morning. Melaya also offers a visual builder for agent workflows, with schedules, connected tools and approval steps.

A founder could build a daily research workflow that gathers a shortlist and prepares outreach drafts for review. An operations team could schedule a recurring check instead of manually repeating the same request. The assistant can work with Melaya's pipelines through MCP, while autonomous workflows run through Melaya's runner.

That combination is the interesting part: an assistant for the work in front of you, and reusable workflows for the work that keeps coming back.

Give the assistant access, not a blank cheque

Melaya makes permissions part of the product. Users choose which capabilities to grant, which browser tabs to attach and which device apps to allow. Teams can configure human approval before consequential actions and inspect execution history.

For someone connecting an assistant to business systems or a personal device, this matters as much as the tool list. The useful question is not just whether an agent can act, but where it can act and when a person stays in control.

It also shaped the observability integration. Melaya wanted to improve the experience without exporting raw tool inputs, outputs, phone screens or browser page contents to Flowlines.

The first product question was account readiness

Once those capabilities reach an assistant, the next question is what people actually try. Melaya reports that roughly 49% of calls in the initial window included optional client-supplied intent context. Flowlines grouped activity around recognizable jobs: checking the connection, understanding available capabilities, and reviewing account readiness.

That finding is more useful than a total call count. Before asking an assistant to carry out a complex workflow, people were checking whether their account and connected services were ready. Melaya already had a setup check for that job. The observed pattern made its role in onboarding clearer.

For Melaya, that gives onboarding a clear focus: help people confirm that their account is ready, then make the first useful action easy to find.

See the tools people reach for

Across 754 calls, 44 distinct tools were used. Flowlines gave Melaya a first map of the capabilities people were reaching for.

The figures come from Melaya's September 2026 integration review. The first 48-hour window includes internal validation traffic and deliberate failure tests: an early baseline, not an adoption benchmark or a measured uplift.

Two days without a call does not make a tool unnecessary. It may serve an infrequent job, depend on a permission most users have not granted, or be hard for a client to discover. Conversely, a busy tool may be helping people complete work, or it may be receiving repeated attempts.

This is where MCP analytics becomes a product tool. Review the attempted job alongside the sequence of calls. Separate setup checks from ongoing use. Look for repeat usage rather than treating every call as another successful customer interaction.

For Melaya, the breadth of the product makes this especially valuable. Browser work, device actions and agent workflows should not all be judged by the same raw activity number.

Know what moved between releases

The observation window spanned three Melaya releases. Once release context was included in telemetry, Flowlines could associate calls with the version that served them.

The review also describes detected input and output schema changes on two pipeline tools. This matters because the tool description and schema are part of the interface an AI client uses to decide what to call. Changing that interface deserves attention even when the server remains available.

Melaya can use that signal to review the affected tools and compare behavior across releases, instead of waiting for someone to report that an assistant got stuck. A schema change is a reason to check, not by itself proof of a regression.

Count tool outcomes without confusing them with user outcomes

Melaya reported 732 successful calls and 22 failed calls, with no unknown call statuses. That works out to a 97.1% tool-call success rate in this window.

It does not mean 97.1% of people completed their task. A successful call can still be one step in an unfinished workflow, and a failed call can be followed by a successful recovery.

Keeping those distinctions visible makes the baseline useful. The team can investigate failures without inflating customer impact, and examine tool usage without presenting it as verified business value. For a deeper framework, see our MCP observability guide.

Keep the export boundary explicit

Melaya chose not to export raw tool arguments, results, phone screens or browser page contents to Flowlines. It also excludes direct identifiers such as names and email addresses. Tool identity, outcomes, release and client context provide the operational view; a pseudonymous account identifier supports account-level analysis.

Pseudonymous identifiers are not anonymous data, and optional client-supplied intent summaries can still contain sensitive information. Both need care. Excluding payloads also limits what can be established about content quality and final outcomes. Within that boundary, Melaya can still learn about tool adoption, common jobs and release changes.

A partnership that improved both products

The integration gave Melaya a clearer way to read its MCP usage. It also surfaced practical gaps: calls needed release context, and stateless traffic needed a useful journey view.

Melaya added release information and reduced the overhead of collecting optional task context. Its feedback helped Flowlines improve journey grouping for stateless MCP traffic and clarify setup guidance. Those inferred journeys support investigation rather than guaranteeing a reconstruction of the original conversation.

Melaya brought more than a server to monitor. Its team tested the integration against real execution constraints and helped improve the product. Flowlines brought a way to turn that activity into decisions about onboarding, tools and releases.

Two useful places to start

Want your assistant to work across your browser, Android apps and business tools? Explore the Melaya MCP server ↗. Choose the capabilities and permissions that fit your workflow, and follow Melaya's setup requirements for each surface.

Building an MCP server yourself? Connect Flowlines, then let your assistant guide the separate instrumentation step with the official Flowlines plugin ↗. Start with a reviewed export boundary and a representative journey. The Free plan includes 100 analyzed MCP sessions per month, without a credit card.

Read the Melaya partner overview for the two products side by side, or use the MCP instrumentation guide to plan your own setup.