MCP Guide

Building Outli.ne profiles with an AI agent

Outli.ne exposes a remote Model Context Protocol server so an AI agent — running on your own Claude (or any MCP client) — can create, manage, and optimize your profile. You curate; the agent builds. Outli.ne runs no inference of its own.

What do you want to do?

Pick your goal — each workflow is a copy-pasteable prompt ready to paste into Claude.

Keep it current →

You've been posting — sync your latest work to your profile.

Fill it out →

Your profile is sparse. Add the sections and cards it's missing.

Add something new →

New offer, service, course, or booking link to announce.

Optimize it →

See what's working, find the weak spot, fix it.

Go live →

Check the profile is ready, then publish.

Get started in 3 steps

Mint a PAT

Go to Dashboard → MCP tokens and create a Personal Access Token. Copy it — it's shown once.

Add the server to Claude

Drop the config snippet into your claude_desktop_config.json or Claude Code settings. Point it at https://outli.ne/api/mcp with your token as a Bearer header.

Say "build my profile"

Open a new Claude conversation and say: “Build out my Outli.ne profile.” The agent calls get_guidance first and follows the recommended loop automatically.

Example prompts to try

“Build out my Outli.ne profile”
“How's my profile doing this week?”
“Import my last 10 Instagram posts”
“Publish my profile”
“What card kinds can I use for free?”

Using this page in Claude?

Copy the URL https://outli.ne/docs/mcp and paste it into Claude with a message like “Read this page and help me connect Claude to Outli.ne.” — the guide is plain text and Claude can follow it directly.

Overview

The Outli.ne MCP server is a remote Streamable HTTP endpoint, owner-scoped per token. The server handles no inference — Claude (running on your account) is the brain; Outli.ne provides typed, permissioned tools that write directly to your profile. On connect, the server returns an in-band instructions block that orients the agent to the recommended loop and the cardinal rules, so you don't have to paste a system prompt.

Authentication

1. Mint a Personal Access Token (PAT)

Go to Dashboard → MCP tokens at outli.ne/dashboard.
Create a token with a label you'll recognize (e.g. “Claude Desktop”).
Copy it once — it looks like oln_pat_xxxxxxxx…. It is shown only at creation; stored hashed and cannot be retrieved later. Revoke and re-mint if you lose it.

A token is scoped to your account only. Every tool call operates on your own outline(s); there is no cross-account access, even for admins.

2. Connect the server

The MCP server URL:

bash

https://outli.ne/api/mcp

Authenticate with a Bearer header:

http

Authorization: Bearer oln_pat_xxxxxxxx…

Claude Desktop / Claude Code config

json

{
  "mcpServers": {
    "outline": {
      "type": "http",
      "url": "https://outli.ne/api/mcp",
      "headers": { "Authorization": "Bearer oln_pat_xxxxxxxx…" }
    }
  }
}

On connect, the server returns an instructions block orienting the agent to the recommended loop and the cardinal rules. You don't need to paste a system prompt — it's in-band.

How Claude builds your profile

Tell your agent: “Build out my Outli.ne profile” and it should follow this order. Each step maps to a tool:

Step	Tool	What it does
1	get_guidance	Outli.ne's own playbook: goal→structure templates, section grammar by creator type, the free vs Pro card-kind catalogue. Start here.
2	get_taste_fingerprint	A compact summary of how you've curated before, so the agent matches your taste instead of starting cold.
3	import_from_instagram / import_from_url	Pull raw material (posts, captions, links) into your media library.
4	add_stack · add_card · update_card · reorder_cards	Build the profile to the chosen template.
5	set_social_links · set_seo	Finish identity + discoverability. `set_seo` requires Pro.
6	publish	Go live at your handle URL.
7	get_analytics	Read your last-7-days views, trend, clicks, and intent — then iterate. `log_curation_decision` records what you kept vs rejected so future builds get smarter.

The cardinal rules (the agent is told these too)

Stay in your voice; never invent facts not in your bio, captions, or what you tell it.
Every card should have a photo cover — text-only cards look bare on the full-screen layout.
CTA on the lead card of each stack, not every card.

Tools reference

Read

get_guidance	Playbook, templates, and free/Pro card catalogue.
get_taste_fingerprint	Compact summary of past curation decisions.
get_profile	Full profile structure — stacks, cards, social links, SEO.
list_cards	All cards in a stack, in display order.
get_analytics	7-day views, clicks, intent, and sparkline.

Write

add_stack	Create a new section.
add_card	Add a card to a stack.
update_card	Edit title, body, cover image, kind, or CTA on an existing card.
reorder_cards	Reorder cards within a stack.
set_social_links	Set handle/URL for each social platform.
set_seo	Set OG title, description, and image. Requires Pro.
publish	Make the profile live at the handle URL.

Import

import_from_instagram	Scrape and import posts from a public Instagram handle into the media library.
import_from_url	Import content from any public URL.

Utility

log_curation_decision	Record what you kept or rejected and why — improves future taste fingerprints.
delete_card	Remove a card. Returns a preview unless confirm:true is passed.
delete_stack	Remove a stack. Returns a preview unless confirm:true is passed.

Analytics

get_analytics returns your real engagement for the last 7 days — free for everyone:

json

{
  "outlineId": "42",
  "slug": "yourhandle",
  "window": "7d",
  "stats": {
    "views7d": 120,        // profile pageviews, past 7 days (bots filtered)
    "viewsPrev7d": 80,     // prior 7 days, for the trend
    "trendPct": 50,        // +50% week over week (null if prior week had 0 views)
    "clicks7d": 14,        // link/button + social-icon clicks
    "intent7d": 3,         // booking + purchase intent events
    "sparkline": [10, 12, 18, 20, 22, 19, 19]  // daily views, oldest → newest
  }
}

If your profile isn't published yet, or analytics aren't configured, stats is null and a note explains the next step — the tool never fabricates numbers.

Ask things like “How's my profile doing this week, and which stack should I rework?” — the agent reads get_analytics, cross-references the structure with get_guidance, and proposes changes.

Rate limits

Calls are rate-limited per account — generous defaults designed for abuse prevention, not throttling normal use:

Bucket	Limit	Tools
Reads	300 / hour	get_guidance, get_analytics, get_profile, list_cards, get_taste_fingerprint, …
Writes	120 / hour	add_card, update_card, publish, set_seo, …
Instagram imports	20 / hour	import_from_instagram
Token creation	20 / hour	(dashboard)

Safety notes

Destructive tools (delete_card, delete_stack) return a preview unless you pass confirm:true, and refuse to remove the last card/stack.
Conversion card kinds (contact, book-me, embed-booking, buy, email) and set_seo require a Pro plan; the tool returns a clear FORBIDDEN_FEATURE error if you're not entitled.
Every tool is owner-scoped — a token can only ever touch its own account's profiles.

Ready to connect? Generate your access token from the dashboard.

Go to Dashboard → Connect