Teach Any AI Agent to Work with Email: Introducing the InboxParse Skill

AI coding agents are increasingly capable — but when it comes to email, most of them draw a blank. They can write code that calls an API, but they don't know your API unless you teach them.

Today we're publishing the InboxParse Agent Skill — a single install that gives any Vercel-compatible AI agent (Claude Code, v0, Cursor, and others) deep knowledge of the InboxParse Email-to-LLM API. One command, and your agent knows every endpoint, authentication pattern, and workflow.

What Are Agent Skills?

Agent Skills are an open ecosystem introduced by Vercel. A skill is a structured set of instructions — packaged as a GitHub repository — that teaches an AI agent how to work with a specific tool or API. When installed, the agent reads the skill's SKILL.md file and activates it whenever the user's request matches the skill's trigger phrases.

Think of it as a README that your AI agent actually reads and follows. No SDK, no plugin framework — just Markdown files that give the agent the context it needs to help you.

Install in One Command

npx skills add inboxparse/skill

That's it. The skill lands in your project's skills/ directory and is automatically picked up by compatible agents. You can also browse it on skills.sh.

What's Inside

skill/
├── SKILL.md                        # Main skill instructions (loaded by agents)
├── references/
│   ├── api-reference.md            # Full V1 API endpoint reference
│   ├── webhook-events.md           # Webhook event types and payloads
│   └── error-codes.md              # Error codes and troubleshooting
└── assets/examples/
    ├── list-emails.sh              # curl examples
    ├── search-emails.sh
    ├── send-email.sh
    └── setup-webhook.sh

The core is SKILL.md — a frontmatter-annotated Markdown file that tells the agent what InboxParse does, when to activate, and how to use every endpoint:

---
name: inboxparse
description: >
  Work with the InboxParse Email-to-LLM API. Use when the user mentions
  "InboxParse", "email API", "parse emails", "email-to-LLM", "email parsing",
  or wants to integrate email data into AI workflows.
metadata:
  version: "1.0.0"
  author: "InboxParse"
  homepage: "https://inboxparse.com"
license: MIT
---

The references/ directory contains detailed API docs, webhook event types, and error codes. The agent pulls these in when it needs deeper context — like looking up the exact payload shape for a webhook event or checking what a 409 error means.

The assets/examples/ directory has ready-to-run curl scripts the agent can reference or suggest to the user.

20+ Endpoints, One Skill

The skill covers the complete InboxParse V1 API:

Included Examples

The skill ships with curl scripts your agent can reference. Here are a few:

List recent emails

#!/bin/bash
# List recent emails in markdown format
API_KEY="${INBOXPARSE_API_KEY:?Set INBOXPARSE_API_KEY}"
curl -s -H "Authorization: Bearer $API_KEY" \
  "https://inboxparse.com/api/v1/emails?limit=10&format=markdown" | jq .

Search with hybrid mode

#!/bin/bash
# Search emails with hybrid mode
API_KEY="${INBOXPARSE_API_KEY:?Set INBOXPARSE_API_KEY}"
QUERY="${1:?Usage: search-emails.sh <query>}"
curl -s -X POST -H "Authorization: Bearer $API_KEY" \
  -H "Content-Type: application/json" \
  -d "{\"query\": \"$QUERY\", \"mode\": \"hybrid\"}" \
  "https://inboxparse.com/api/v1/search" | jq .

Set up a webhook

#!/bin/bash
# Create a webhook for new emails (requires admin key)
API_KEY="${INBOXPARSE_API_KEY:?Set INBOXPARSE_API_KEY}"
WEBHOOK_URL="${1:?Usage: setup-webhook.sh <url>}"
curl -s -X POST -H "Authorization: Bearer $API_KEY" \
  -H "Content-Type: application/json" \
  -d "{\"url\": \"$WEBHOOK_URL\", \"events\": [\"email.received\", \"email.ai_processed\"]}" \
  "https://inboxparse.com/api/v1/webhooks" | jq .

How It Works with Your Agent

When you mention "InboxParse", "email API", "parse emails", or related phrases, the agent activates the skill and loads the full API reference into its context. Here's what a typical interaction looks like:

User: I need to search my InboxParse emails for overdue invoices

Agent activates the InboxParse skill, reads SKILL.md, and responds:

Agent: I'll use the search endpoint with hybrid mode to find those.
Here's the API call:

  POST https://inboxparse.com/api/v1/search
  Authorization: Bearer ip_<your_key>
  Body: { "query": "invoice overdue", "mode": "hybrid", "limit": 20 }

Hybrid mode combines full-text and semantic search, so it'll
catch invoices described as "past due" or "payment outstanding"
even if the word "overdue" isn't used.

The response includes AI-generated summaries and labels for
each match, so you can triage results without opening every email.

The agent doesn't just parrot docs — it combines its knowledge of the API with your specific request to give targeted guidance. It knows about authentication (the ip_ key prefix, member vs admin roles), pagination (cursor-based), response formats (markdown, full, raw), and error codes.

Skill vs. MCP Server

InboxParse offers two ways to connect AI agents — and they serve different purposes:

Use the skill when you're building something — the agent helps you write correct API calls. Use the MCP server when you want the agent to interact with your email directly.

Get Started

Install the skill and start building with email data in under a minute:

1. Run npx skills add inboxparse/skill in your project.
2. Ask your agent something like "How do I list emails with InboxParse?" — it should activate the skill and guide you through the API.
3. Grab an API key from Console → API Keys and start making requests.

• GitHub repo
• View on skills.sh
• Create a free InboxParse account
• Full API documentation