AI Prompts for MCP API Connectors
Free tested AI prompts for MCP API Connectors. Built for real results you can use right away.
Free AI prompts for MCP API Connectors, tested and ready to use right now.
AI Prompts for MCP API Connectors
Free tested AI prompts for MCP API Connectors. Built for real results you can use right away.
Browse top AI prompts for MCP API Connectors across build connectors for your stack, and more. Every prompt in this guide is free to copy and built for real results. No prompt engineering experience needed.
Stage 1
Build Connectors for Your Stack
The right connector does the minimum required. Every unnecessary capability is a capability that can fail or be misused. These prompts help you build connectors that are scoped to exactly what your loop needs.
Build a Linear MCP connector
Write the complete MCP connector implementation for Linear for [YOUR_PROJECT]. The connector should expose: list_issues (filtered by project and status), get_issue (full issue with comments), create_issue (with title, description, and label), update_issue_status (move to in-progress or done), and add_comment (post a comment on an issue). Write the full TypeScript implementation using the @modelcontextprotocol/sdk and the Linear GraphQL API, including authentication via environment variable.
Build a Slack notification connector
Write an MCP connector for Slack that my agent loop uses to send status updates for [YOUR_PROJECT]. The connector should expose: post_message (accepts channel and text), post_formatted_message (accepts channel and a structured payload with header, fields, and color), and list_channels (returns available channels the bot has access to). Write the implementation using the Slack Web API and incoming webhooks.
Build a GitHub connector for loop actions
Write an MCP connector for GitHub that supports the full PR lifecycle for [YOUR_PROJECT]. The connector should expose: list_prs (filtered by state and label), get_pr_diff (returns the full diff for a PR number), create_pr (opens a PR with title, body, base branch, and head branch), post_review (posts a review with a verdict and line comments), add_label (adds a label to a PR or issue), and merge_pr (merges a PR that has passed all checks). Write the TypeScript implementation.
Build a Supabase read connector
Write a read-only MCP connector for Supabase for [YOUR_PROJECT]. The connector should expose: query_table (accepts table name, optional filters, and limit), get_record (accepts table name and row ID), and run_rpc (calls a Supabase RPC function by name with parameters). Include: authentication using the service role key from environment variables, a whitelist of allowed tables passed in at server startup, and row-level sanitization that removes password and token fields from all responses.
Build a Notion connector for agent documentation
Write an MCP connector for Notion that my agent uses to update project documentation in [YOUR_PROJECT]. The connector should expose: get_page (returns the page content as markdown given a page ID), update_page (replaces the content of a page), append_to_page (adds a block to the end of a page without replacing existing content), and create_page (creates a new page in a specified database). Write the implementation using the Notion API.
Frequently asked questions
What is the difference between an MCP server and an MCP connector?+
An MCP server is the process that runs and exposes tools. An MCP connector is a specific integration within that server, such as a GitHub connector or a Slack connector. You can run one MCP server that hosts multiple connectors, each connecting to a different external service. Most loop projects need one server with two to four connectors covering the tools the agent needs to act on.
How do I give an agent safe access to my production database?+
Start with a read-only connector that uses a service account with SELECT permissions only. Add a whitelist of allowed tables at server startup so the agent cannot query anything outside that list. Strip sensitive fields like passwords, tokens, and API keys from every response before returning it. Only add write tools after the read-only connector is tested and stable, and give each write tool a dry_run parameter.
Which APIs should I prioritize connecting to my agent loop?+
Start with the tools your loop already uses manually: usually GitHub for reading issues and opening PRs, and a notification channel like Slack so you know what the loop did each run. Add database access only when the loop needs to read data to make decisions. Add ticket creation tools like Linear only when the loop is stable enough to create real records you want to track.
How do I handle authentication in an MCP connector?+
Store all credentials as environment variables and read them at server startup, never at the point of each tool call. The server should fail to start if required credentials are missing rather than failing silently on the first tool call. Never log credential values in the server output. The connector implementations in Stage 1 of this guide follow this pattern for every external service.
Can the same MCP connector be used by multiple loops?+
Yes. An MCP server is just a process that exposes tools via the protocol. Multiple Claude Code sessions or loop instances can connect to the same server simultaneously. The only risk is write contention: if two loops both try to create a ticket or merge a PR at the same time. Rate limiting at the connector level and task scoping at the loop level both prevent this.
Other prompt packages you might like
AI Prompts for MCP Server Setup
A loop that can only see the filesystem is a tiny loop.
See prompts
AI Prompts for Loop Engineering
Loop engineering is what happens when you stop prompting the coding agent yourself and design the system that prompts it instead.
See promptsAI Prompts for Claude Code Loop Setup
Claude Code ships all five loop engineering primitives: scheduled tasks, worktrees, sub-agents, skills, and MCP connectors.
See prompts