Connecting an agent to Cabgo's MCP solves access: from that point on, the agent can check operation status, review driver metrics, update configurations and execute actions with two-step confirmation. What it doesn't solve is the consistency of daily output. Operators who get the most from their agent aren't the ones who ask more elaborate questions — they're the ones who keep a small set of stored prompts that produce exactly the output they need, in the format they can use without editing. Without that set, every session starts from scratch: an improvised question, a generic response, and a result that needs reformatting before it can be shared. With it, the agent and operator work from the same template every day, making today's output directly comparable to last week's.

This article documents five high-utility prompts for regional ride-hailing operators with 40 to 150 active drivers who already have Cabgo's MCP connected to their agent — Claude Desktop, Claude Code, ChatGPT or Cursor. For each prompt we explain when to use it, which MCP server tools it activates, and what response format works best. The goal isn't for the operator to memorize the templates: it's for them to save them in the agent client as custom instructions, or in a text file they can copy and paste at the start of any working session.

Why the generic prompt produces output you can't reuse

An agent connected to an MCP responds to the input it receives. When that input is an open question — "how did the day go?" — the agent infers which dimension of "how did it go" is relevant: active drivers? completed trips? cancellations? estimated revenue? The inference is reasonable, but every time the operator asks the same question in slightly different terms, the agent structures the output differently. Monday's result doesn't have the same format as Thursday's, which makes it impossible to compare them at a glance. The operator who wants to know whether active drivers this week are up or down from last week has to read two narratives instead of comparing two numbers in the same position.

The antidote isn't to write rigid prompts — it's to write prompts with defined structure. A well-designed prompt specifies two things: what dimensions the agent should cover and what format it should use to present the result. With that, the agent knows exactly which MCP tools to call without follow-up questions, and the output has the same structure every day. The operator can read it in 60 seconds and file it for comparison next week.

Prompt 1 — Shift summary: operation status in 60 seconds

The shift summary is the most commonly used prompt in operations with 60 to 130 drivers. It covers the state at the end of the day or shift: how many drivers were active, how many trips were completed, what the cancellation rate was, and which three drivers had the most trips and which three had the fewest. The operator who has this prompt saved pastes it when opening the chat with the agent each afternoon and has the day's status in two minutes without needing to open any dashboard. The agent activates cabgo_my_app_status and cabgo_driver_activity from the MCP server to build the full picture without the operator specifying the tools.

Prompt 2 — Inactive drivers: who hasn't worked and for how long

Driver inactivity is one of the earliest signals that something is changing — it could be an unresolved complaint, an offer from another platform, or simply an access problem with the app. In operations with more than 60 registered drivers, the coordinator doesn't catch inactivity through manual review. A specific prompt gives the agent clear instructions to query activity history through cabgo_driver_activity and return a list sorted by days without trips, which the coordinator can use directly for follow-up. This prompt should run once per week, not daily: meaningful inactivity doesn't change in 24 hours.

Prompt 3 — Payment mix: cash control per driver in a single view

Cash control is the most frequent operational problem in LATAM fleets, and the one that generates the most invisible losses when not measured regularly. A payment mix review prompt asks the agent to query each driver's outstanding balance and classify which ones are above the defined threshold. The key to making this prompt work well is including three alert zones in the text itself: green for the normal range, yellow near the limit, and red when the balance requires immediate coordinator action. With that framework, the output is directly actionable — the coordinator doesn't have to interpret the numbers, just act on those in the red zone.

Prompt 4 — Communication draft: driver notices without starting from scratch

Drafting a notice for drivers — about a fare change, a new operating zone, a high-demand event, or a policy update — is a task the operator does 4 to 8 times per month. Without a saved prompt, that drafting consumes 15 to 30 minutes of back-and-forth with the agent adjusting tone, length, and format. With a saved template, the agent produces the draft on the first response and the operator only needs to verify the specific data before sending. The key to this prompt isn't what you ask the agent — it's how much fixed context you give it: the team's tone, the output channel, and the exact format you expect.

Prompt 5 — Weekly close: key metrics in a shareable format

The weekly close is the report most operators want to produce consistently but fewest manage to sustain. Without a specific prompt, generating it takes 30 to 60 minutes of manual work cross-referencing dashboard data. With a well-designed prompt, the agent queries the MCP server, structures the summary, and produces a text the operator can copy directly into an email, a WhatsApp group with the team, or a shared document with partners. This is the longest of the five prompts — because the output it produces is also the most complete.

How to build and maintain your own prompt library

The five prompts above are starting points, not closed formulas. Every operation has its own particulars — a high-demand zone that doesn't appear in standard reports, a driver category that requires a specific filter, or a communication format the team already recognizes. The operator who adapts these templates to their context and saves them in the agent client has a library that improves with every week of use. The signal that a prompt needs revision is simple: if the output it returns requires more than 30 seconds of editing before it can be used, the prompt is asking for more specificity.

The difference between an agent an operator uses every day and one they open only when they remember it exists isn't in the agent — it's in whether the operator has built a routine around it. That routine starts with a small set of prompts that produce consistent output: enough to make the daily review take 5 minutes instead of 30, and to ensure the weekly close is no longer a task that gets postponed indefinitely. The five prompts in this article cover the most frequent use cases in mid-size operations. Four weeks of consistent use is enough for the operator to know which ones to adapt and which new ones to add to their library.

The agent connected to Cabgo's MCP has access to live operation data — active drivers, outstanding balances, shift trips, ratings, payment mix. That data exists on the server whether or not the operator has saved prompts. The difference is that with well-designed prompts, that data becomes a summary that can be read, shared, and acted on in the first minutes of the day. Without them, the data is still available — but it requires a new conversation every time to extract it in the right format. The prompt library doesn't make the agent smarter: it makes the operator more efficient.