Corcava logoThe Only Business Tool You NeedCorcava
Menu
Solutions
Industries
DesignersHR RecruitingLaw FirmsReal EstateConsultingDigital MarketingAccountingEcommerceSoftware Development
Product
FeaturesIntegrations
BlogDownload
Resources
TemplatesDocsMCPFAQAPIComparisonsRoadmapWhat's New$ Affiliate Program
Pricing
Free 14‑day trial

No credit card required

Prompting Patterns for MCP Tools: Read-First, Summarize, Then Act

Master the art of prompting AI assistants that have MCP tools. This patterns guide teaches you how to ask for plans, preview diffs, use confirmation tokens, and batch operations—all with Corcava-specific examples you can use immediately.

Core Prompting Patterns

Pattern 1: Ask for a Plan

Before taking action, ask the AI to create a plan:

Plan-First Pattern

"Create a plan for updating all tasks in project '[Project ID]' to status 'in_progress'. Show me the plan first, including which tasks will be updated. Don't update anything yet—just show the plan."

What the AI does:

  1. Calls list_tasks to get tasks in project
  2. Creates a plan showing which tasks will be updated
  3. Shows you the plan for review
  4. Waits for approval before updating

Benefits: Review before action, catch mistakes early, understand scope

Pattern 2: Preview Diffs

Show what will change before applying updates:

Diff Preview Pattern

"Update task '[Task ID]' status from 'open' to 'in_progress'. First get the current task, then show me a diff of what will change. Only update after I approve the diff."

What the AI does:

  1. Calls get_task to get current state
  2. Creates diff showing before → after
  3. Displays diff clearly
  4. Waits for approval before calling update_task

Benefits: See exactly what changes, verify correctness, prevent mistakes

Pattern 3: Confirmation Tokens

Require specific tokens to confirm important operations:

Token Confirmation Pattern

"Create a task in Corcava with title '[Title]' and project '[Project ID]'. Show me a preview. To confirm, I must type the exact token 'CREATE-TASK-2026'. If I don't type this token, don't create anything."

What the AI does:

  1. Shows preview of task to be created
  2. Waits for exact token match
  3. Only calls create_task after token received
  4. Prevents accidental confirmations

Benefits: Prevents accidental yes/no responses, requires intentional confirmation

Pattern 4: Batching

Group operations to reduce tool calls:

Batching Pattern

"List tasks with status 'blocked' and due this week. Then update the top 5 to status 'in_progress'. Show me which tasks will be updated before applying changes."

What the AI does:

  1. Calls list_tasks once with filters
  2. Selects top 5 from results
  3. Shows preview of updates
  4. Batches 5 update_task calls after approval

Benefits: Fewer tool calls, faster execution, better performance

Read-First Pattern

Always read before writing:

Read-First Principle

Before any write operation, always read the current state first:

"Before updating task '[Task ID]', first get the current task details using get_task. Show me the current status, assignee, and due date. Then I'll tell you what to change."

Why: Prevents overwriting data, ensures accuracy, shows what will change

Summarize Then Act Pattern

Have the AI summarize before taking action:

Summarize Then Act

"Summarize what needs to be done: list all blocked tasks, identify why they're blocked, and propose next steps. Show me the summary. Then, after I approve, add comments to each task with the proposed next steps."

What the AI does:

  1. Reads tasks and analyzes blockers
  2. Creates summary with proposed actions
  3. Shows summary for review
  4. Only acts after approval

Complete Pattern Examples

Example 1: Safe Task Creation

Complete Safe Create Pattern

"Create a task in Corcava: - Title: Review API documentation - Project: [Project ID] - Due date: 2026-02-01 1. Show me a preview of what will be created 2. Show me a diff (empty → new task fields) 3. Wait for me to type the confirmation token 'CREATE-TASK-2026' 4. Only after I type the exact token, create the task 5. If I don't type the token, don't create anything"

This combines: Preview + Diff + Confirmation Token

Example 2: Batch Update with Approval

Complete Batch Update Pattern

"List tasks in project '[Project ID]' with status 'open' and due this week. Then: 1. Show me the list of tasks 2. Propose updating the top 5 to status 'in_progress' 3. Show me a diff for each task (status: open → in_progress) 4. Wait for me to type 'APPROVE BATCH UPDATE' 5. Only then update all 5 tasks 6. If I don't approve, don't update anything"

This combines: List → Select → Preview → Diff → Batch Update

Pattern Combinations

Combining Patterns

You can combine multiple patterns for maximum safety:

  • Read-First + Diff + Token: Get current state → Show diff → Require token
  • Plan + Preview + Approval: Create plan → Show preview → Get approval
  • Batch + Summarize + Confirm: Batch operations → Summarize → Confirm all

Best Practices

Prompting Best Practices

  • Always read first: Get current state before updating
  • Show previews: Always preview before write operations
  • Use confirmation tokens: For critical operations, require specific tokens
  • Batch when possible: Group related operations together
  • Be explicit: Clearly state what should and shouldn't happen
  • Specify output format: Tell AI how you want results formatted

Related Resources

Write Approval Patterns

Deep dive on approvals

50 Prompt Pack

Ready-to-use prompts

Batching Guide

Reduce tool calls

Hallucination Prevention

Verify before writing

Master MCP Prompting Patterns

Use these patterns to prompt AI assistants with MCP tools effectively and safely

50 Prompt Pack →Write Approval