MCP Example Workflows

Sample AI assistant prompts for daily operations, period close, investigation, cleanup, deliverables, onboarding, and reusable MCP routines.

Tip: Default to read-only. When a workflow writes data, ask for a preview first and confirm before applying. The prompts in this article are phrased that way; keep that pattern when adapting them.

How to read this page

Every workflow is tagged in its title so you know what it touches before you open it:

• Read-only: looks things up, never changes Kick.
• Read first, then writes: starts as a lookup; writes only after you approve a preview.
• Writes to Kick: uses a write tool. Always preview and confirm before applying.

A label that says "multi-step" spans several prompts or turns, or builds a reusable artifact.

Hover a prompt and click the copy icon to paste it into your assistant. Use Ask → Open in Claude or Open in ChatGPT (top right) to load the full page as context.

You can ask the assistant to do more than run one prompt. It can chain steps, build a reusable artifact (a checklist, a Q&A sheet), save a sequence as a routine, or draft a skill. See Routines & artifacts below.

Start here

New to Kick MCP? Begin with these read-only single prompts, then explore the sections below.

• Find the right client context: confirm which workspace you are in (see Daily ops).
• Pull reports for review: get a P&L or balance sheet (see Close).
• Review transactions that need attention: triage a period (see Daily ops).

Daily ops

Workspace context snapshot (read-only)

Give me a one-screen snapshot of the Acme LLC workspace: entities, connected accounts, uncategorized count, unmatched transfers, open tasks, and most recent activity. Confirm this is the right workspace before we proceed.

A faster orientation than listing workspaces by name. Use it at the start of a session.

Find the right client context (read-only)

List the Kick workspaces I can access, then help me find the one for Acme LLC and confirm we are in the right place before using any other tools.

Use this when you are switching between clients or checking whether the assistant is using the right workspace.

Daily triage across clients (read-only)

Across all my workspaces, which clients are at risk of missing close this month? Show the workspace, the uncategorized count, the unmatched-transfer count, and the date of the most recent meaningful change.

For a full morning standup view that builds a reusable artifact, see Multi-client morning briefing.

Review transactions that need attention (read-only)

In the active workspace, find January 2026 transactions that need bookkeeping review. Limit the first result to 25 rows. Group your summary by likely follow-up: missing category, unclear counterparty, possible transfer, or needs client context.

Ask for source data before acting:

For each item, include the transaction id, date, amount, counterparty, and bank description.

Account health check (across workspaces) (read-only)

Across every workspace I can access, list connected financial accounts and flag any with no transactions in the last 14 days or a sharp volume drop vs. the prior month. Return one table: workspace, account, last-activity signal, this-month vs. last-month count, sorted most stale first.

Use this to catch broken or stale bank feeds before they block close.

Document backlog (read first, then writes)

Start read-only by finding documents that are not yet linked:

In the Acme LLC workspace, list documents uploaded in the last 30 days that aren't yet matched to a transaction. Group by document category, entity, and account, and flag any that aren't reconciled. Include the file name, upload date, and period for each.

Then match the oldest, and stop before linking:

For the five oldest unmatched documents, use each document's period and account to search transactions in that account within the period (±5% on any extracted amount), rank candidate matches, and stop before linking.

Note: filename search does not read document contents. Match on period, account, and amount, not vendor text.

Task follow-up queue (read first, then writes)

List open tasks in the Acme LLC workspace due this week. For each bookkeeping-related task, summarize the underlying issue and whether the related transactions are still unresolved.

Then stage a new task and stop before confirming:

Prepare a task preview: "Follow up with client on Stripe: $1,240 on Jan 15" linked to transaction T-4821. Stop before confirming.

Close

Pull reports for review (read-only)

Pull the January 2026 P&L for entity 123 on a cash basis. Summarize the largest expense categories and list three questions I should ask the client.

For a deeper review:

Compare the January 2026 P&L, balance sheet, and trial balance for entity 123. Point out accounts that changed materially from the previous month and include the report source for each observation.

Run a period-end sweep (read-only)

A sweep stitches together close-prep checks into a single read-only pass:

For the Acme LLC workspace and the period Jan 1 through Jan 31, 2026, list: every uncategorized transaction, every unmatched transfer, every transaction over $5,000 without a memo, and any vendor that first appeared this period. Group the results and include source IDs.

To reconcile against an external source:

Compare the opening balances in the Acme LLC workspace against the attached trial balance. List every account where the balances diverge, with both values shown side by side. Do not change any data.

For a re-runnable checklist artifact, see Monthly close checklist artifact.

Cash flow vs. P&L sanity check (read-only)

For entity 123 and Q1 2026, compare net income (P&L) to operating cash flow. List the biggest reconciling items (AR/AP, transfers, owner draws, loan principal).

Investigation

Inspect activity before changing anything (read-only)

Show recent activity related to transaction category changes. Summarize who made the changes, when they happened, and whether any batch looks worth reviewing before close.

Use this before approving a close, reviewing a catch-up file, or investigating unexpected report changes.

Unwind a change after review (writes to Kick, multi-step)

When activity inspection turns up a batch you want to roll back, ask for a preview before reverting:

Pull the activity log for the bulk update on Jan 12 that touched 84 transactions. Show what changed for each: the before and after state. Then prepare to revert that batch and stop before applying. I will confirm.

For a single transaction:

Show the full edit history for transaction T-4821. List every change with the user, timestamp, and the previous and new values for each field.

Why did Kick categorize this? (read-only)

For transaction T-4821, explain how it got its current category, class, and counterparty. Check applied rules, similar transactions, and recent edits. List what would change if I recategorized it to Office Supplies.

Account drill-down after a report surprise (read-only)

P&L shows Software Subscriptions up 60% month over month. Pull account transactions for January 2026, group by counterparty, identify the three largest drivers, and for each tell me whether it came from a bank import, a rule, or a manual JE.

For a packaged one-page write-up, see Report-surprise investigation artifact.

Cleanup & scale

Audit and tune the Chart of Accounts (read first, then writes)

Start with usage before making changes:

In the active workspace, list each account and category with the number of transactions posted to it in the last ninety days. Flag entries with zero recent usage and note whether each looks safe to disable.

For a structural change, request a preview before any write:

Prepare to create a new Long-Term Liability account named SBA Loan and reclassify the historical loan transactions for First National Bank into it. Show the count of affected transactions, the date range, and the proposed account hierarchy. Stop before writing.

Clean up counterparties and classes (read first, then writes)

Identify duplicates and underused entries before merging or retiring:

List counterparties in the active workspace that look like duplicates of Amazon. Show the variants, the transaction count for each, and the merge target you would pick. Do not merge yet.

Which classes have fewer than five transactions this year? List each with its transaction count and propose which ones to retire.

Once you have reviewed the proposal, authorize the merge:

Prepare to merge the counterparty variants you listed above into a single counterparty named Amazon. Show the final transaction count and stop before confirming.

Build and tune rules (read first, then writes)

Preview what a new rule would match before creating it:

I want a rule that categorizes Gusto payouts as Payroll and assigns class = Operations. Show me which existing transactions would have matched if this rule had been in place. Stop before creating the rule.

Audit existing rules for staleness:

List my existing rules with the date each one last fired. Flag any that have not fired in the last six months and check whether the underlying patterns still appear in recent transactions.

Normalize a messy vendor (read first, then writes)

Search transactions containing "AMZN MKTP" in January 2026. Show distinct description variants and current categories. Propose a single counterparty + category mapping and count how many rows a bulk update would touch. Stop before updating.

Review and fix split transactions (read first, then writes)

List transactions in January 2026 split across multiple categories. Flag splits where line amounts don't sum to the parent or where a line lacks a class. Include transaction ids.

Then re-split a specific transaction, preview only:

Prepare to re-split transaction T-4821 as $400 to Software and $100 to Office Supplies. Preview only.

Deliverables

Prepare client follow-up (read-only)

Review the open transaction questions and draft a client-friendly message. Do not change any Kick data. Include transaction dates, amounts, and short context for each question.

This is a strong read-only workflow: the assistant gathers context while the accountant stays in control of the client message.

Owner-facing monthly summary (read-only)

For entity 123 and January 2026, draft a short owner summary: cash in/out, top 5 expenses, largest unusual transactions, and three plain-English questions. Do not change any data.

Vendor spend / 1099 prep (read-only)

Pull expenses by vendor for entity 123 for 2025. Flag vendors over $600 with missing or inconsistent counterparty names. List the top 10 vendors by spend with totals.

Then drill into one vendor:

For vendor Stripe, drill into account transactions and list payments missing 1099-relevant detail.

Multi-entity comparison (read-only)

Compare Acme LLC and Beta Co for Q1 2026: revenue, top expense categories, uncategorized count, and unmatched-transfer count. Call out material differences.

Onboarding

Stage a new client setup (writes to Kick, multi-step)

Onboarding involves writes: entities, addresses, ledgers, Chart of Accounts copy, and opening balances. Stage each step as a preview before confirming.

Prepare to set up a new entity called Beta Co in California. List the steps you will run: entity creation, address, tax locations, ledger setup, Chart of Accounts copy from Acme LLC, and opening-balance staging from the attached trial balance. Then stop before executing. I will confirm each step before you run it.

For comparing existing client structures (read-only):

Compare the Charts of Accounts across the workspaces I can access. List accounts that exist in only one workspace and accounts that exist in all of them. Do not modify anything.

Catch-up onboarding plan (read first, then writes)

New client Beta Co just connected accounts with 9 months of history. Draft a staged catch-up plan artifact: phase the work by month, estimate volume per phase from transaction statistics, and list the CoA copy source to use. Don't execute. For each phase, create a task (preview first) so I can assign it.

Routines & artifacts

These chain multiple steps, run across turns, or produce a reusable artifact, routine, or skill.

Monthly close checklist artifact (read-only, multi-step)

For the Acme LLC workspace and January 2026, run a full close-readiness pass: uncategorized, unmatched transfers, transactions over $5,000 without a memo, accounts with material month-over-month swings, and open tasks. Then build an interactive close checklist artifact I can re-run: each row is a check, its current count, the drill-down query behind it, and a status I can tick off.

Make it durable:

Save the exact sequence of MCP calls you just used as a named routine "Monthly Close: Acme LLC" so I can re-run it next period.

Client Q&A loop (read first, then writes)

Find this month's transactions needing client input. For each, gather id, date, amount, account, and current best-guess category. Build a client-ready Q&A artifact: one plain-English question per item, grouped by client, with a blank "client answer" column. Don't change any Kick data. I'll paste answers back next turn.

Next turn, close the loop:

Here are the client's answers. Map each to the right category/class, show me one combined bulk-update preview, and stop before applying.

Report-surprise investigation artifact (read-only, multi-step)

Pull the January 2026 P&L for entity 123. Take the three accounts with the biggest month-over-month change, drill into each account's transactions, identify the top drivers, and attribute each to a bank import, a rule, or a manual JE. Build a one-page "what moved and why" artifact with a sources column.

Multi-client morning briefing (read-only, multi-step)

Across all my workspaces, build a morning-briefing artifact: per client, show uncategorized count, unmatched transfers, open tasks, oldest unmatched document, and most recent meaningful change, sorted by close-blocking risk. Then save it as a routine I can run each morning.

Rule proposal with backtest (read first, then writes)

Look at how I've manually categorized Uber rides over the last 90 days. If a rule would have produced the same result, draft the rule and show me a backtest: which historical transactions it would have matched and where it would have disagreed with what I actually did. Stop before creating it.

Draft journal entries from source documents, then attach to transactions (read first, then writes)

A high-leverage MCP workflow is handing the assistant a document (payroll register, closing statement, amortization schedule, or complex bill) and asking it to draft journal entries, post them after review, and attach the source document to the related transactions. Documents attach to transactions through transactions_document_links_act, not to journal entries. Always have it draft first, never post directly from a document without review.

I have a payroll register for January 2026. (1) Read it and draft the journal entries with full debits, credits, and line detail. (2) On my confirm, post the entries. (3) Upload this document to Kick and link it to the transactions tied to those entries using transactions_document_links_act. Show a preview before posting and before linking, and stop for my approval at each step.

To audit posted entries afterward:

List every manual journal entry posted to Accumulated Depreciation this year. Show date, amount, memo, and the user who posted each.

Turn a recurring review into a skill (read-only, multi-step)

I do this vendor cleanup pass every week. Watch how I work this one, then draft a skill that captures the steps (find similar descriptions → propose counterparty + category mapping → preview bulk update → wait for my confirm). Include the guardrails: always preview, never auto-confirm, cap at 50 rows per batch.

Preview supported write actions (writes to Kick)

Some write tools return a confirmation preview before they change data. Ask the assistant to stop at the preview:

Prepare a preview for the supported correction, then stop. Do not confirm the action until I approve the exact change.

Only confirm after you review the workspace, dates, amounts, and the exact change.