Agent Inbox · New

Run your content business from a chat window.

Heist connects to Claude, ChatGPT, Cursor, and Copilot through MCP. Plan your week, draft a month of posts, update your brand voice, publish to LinkedIn or WordPress — all from a conversation. You never have to open the app.

Start free Watch 90-second demo
What is this?

What is the Agent Inbox, actually?

MCP — Model Context Protocol — is how AI assistants like Claude and ChatGPT connect to outside tools. When you connect Heist through MCP, your AI gets a set of powers: it can read your brand voice, check what's scheduled, draft posts that match your style, queue them in your calendar, and publish them when you say go.

The Agent Inbox is where those actions land. When Claude drafts ten Instagram posts for you, they show up in the inbox. You review them in the web app or from the AI itself — approve, reject, or tweak — and they're live. No context switching. No re-explaining your brand. No dashboard to learn.

You don't have to use these prompts.
The Heist connector works like any Claude or ChatGPT conversation — ask in your own words, change your mind mid-thread, follow up. The 85 tools fire automatically based on what you're asking for. Everything below is just here to spark ideas.
See it happen

Six examples to spark ideas.

These are examples — copy them verbatim, paraphrase them, or invent your own. The AI talks to your 85 Heist tools the same way Claude already talks to you. The conversation is flexible; this page is just here to show you what kinds of work it can do.

Try something like
"Using Heist, give me my daily content briefing."
Get this

A 200-word rundown of what's scheduled today, yesterday's top and bottom performers, and one concrete recommendation for what to do now.

Heist reads your calendar and analytics, then summarizes — you get the takeaway, not the spreadsheet.
Try something like
"Plan my next week. 5 LinkedIn posts and 3 Instagram posts, across my main offer and one story post per platform."
Get this

8 drafts in your Agent Inbox, each grounded in your brand voice, tagged with why it's likely to land.

Heist pulls your brand voice and your top performers, writes the drafts, and hands you the review — all in one conversation.
Try something like
"Audit my brand brain. What's missing? What's contradicting?"
Get this

Gaps, contradictions, and stale facts — with specific proposals to fix each.

Heist cross-checks your profile against your memory facts and recent posts — surfaces the gaps you've missed.
Try something like
"Look at my last 20 posts. Update my voice profile based on what performed."
Get this

A proposed patch to your brand profile, grounded in your actual winners.

Heist doesn't guess — it cites the specific posts it learned from.
Try something like
"Write a 1200-word blog post on why solopreneurs need a content OS. Source: our conversation today."
Get this

A full draft in your blog editor, styled to match your site, ready to publish to WordPress.

Heist's blog generator pulls your voice, audience, and top performers — comes out sounding like you, not a template.
Try something like
"The LinkedIn post from Tuesday did 3x my average engagement. Mark it as a high performer."
Get this

The post is flagged, your Brain learns what about it worked, next week's drafts lean into that pattern.

Closes the learning loop — your Brain gets smarter from real engagement data.
What can it do?

Everything you'd do in the app. Said out loud.

Plan

  • See your content calendar from a conversation
  • Get a 200-word daily briefing on pipeline and performance
  • Generate a full week's content plan grounded in what works

Create

  • Draft social posts for LinkedIn, X, Facebook, Instagram
  • Generate long-form blog posts in your brand voice
  • Create reusable HTML snippets for CTAs, FAQs, related posts
  • Produce campaign variations of a winning post

Organize

  • Add, update, and remove facts in your Brand Brain
  • Update your brand profile, audience personas, and offer summaries
  • Connect URLs and documents as ongoing Brain feeds
  • Rate posts as high/low performers to teach your Brain

Publish

  • Schedule posts across connected social platforms
  • Publish to WordPress with one sentence
  • Reschedule the calendar, fill gaps, pause autopilot
  • Check platform connection status from chat

Every write requires a reason or source. Every action lands in an audit log. You control whether changes apply instantly or queue for review.

Connect in 30 seconds

Three steps. No code. No config files.

Paste this URL
In Claude.ai or ChatGPT: Settings → Connectors → Add custom connector
https://app.heistbrain.com/mcp
Step 1

Open Connectors in your AI chat

In Claude.ai (Pro/Team/Max) or ChatGPT (Pro/Business): click your name → Settings → Connectors → Add custom connector. A small dialog pops up.

Step 2

Paste the URL and approve

Paste https://app.heistbrain.com/mcp into the URL field, click Add. Heist serves an OAuth consent screen — pick which brands the connector can access, click Approve. A "Heist" tool icon appears in your chats with all 85 tools available.

Step 3

Start talking

Try: "What can you do?" — your AI will list every tool. Then: "Plan a 2-week LinkedIn campaign about my core offer."

Advanced: Claude Desktop, Cursor, Windsurf (stdio clients)

Most users should use the URL paste flow above instead — it works in Claude.ai web and ChatGPT and gives you the same 85 tools. This advanced path is only for users who specifically need a local stdio MCP client (Claude Desktop, Cursor, Windsurf, Zed, VS Code GitHub Copilot) — those clients don't speak the OAuth web flow.

  1. Create an API key in Heist → Settings → Account → Advanced. Choose scopes (read + draft + auto_approve is a good start). Key is shown once — copy immediately.
  2. Run the auto-installer: HEIST_EMAIL=you@example.com HEIST_PASSWORD=... npm run connect:mcp — logs in, mints a key, installs the stdio package globally, and writes the right config file for your OS.
  3. Or paste the config snippet manually. Heist's Connectors page gives you a copy-paste block per client.

If you're not sure whether you need this, you don't. The URL paste above works for 99% of people.

More examples

More example prompts to steal or remix.

None of these are required. Steal them word-for-word, paraphrase them, or just talk to the AI the way you'd talk to Claude — the agent figures out which Heist tools to call either way. These are organized by job-to-be-done so you can find one that matches what you're trying to do.

Plan a campaign

Try something like
"Plan a 2-week LinkedIn + X launch campaign for [my product]. 3 posts per week per platform, mix of educational and promotional. Schedule weekday mornings."
Get this

12 drafts grouped under one campaign in one tool call. Auto-scheduled.

Try something like
"Plan a holiday content calendar from Black Friday through New Year. Daily Instagram + 3x weekly LinkedIn. Mix promotional, behind-the-scenes, customer stories."
Get this

35 drafts across two platforms, scheduled across the holiday window.

Try something like
"Plan next week of LinkedIn — 5 thought-leadership pieces, no product pitch. Pull the tone from my last high performer."
Get this

5 voice-matched drafts. Heist reads your top performer and matches its tone automatically.

Repurpose a winner

Try something like
"Take my best LinkedIn post from the last 60 days and create 5 platform-tailored versions for IG, X, FB, TikTok, and a newsletter intro."
Get this

Same idea, five formats, each with the right length and structure for that platform.

Try something like
"Find my most recent blog post and turn it into 8 social posts: 3 LinkedIn, 3 X, 2 Instagram. Tease the blog and link to it."
Get this

8 promotion drafts that each link back to the blog. No manual coordination.

Try something like
"Generate 5 variations of my best-performing post from the last 90 days — different hooks, same core message. I want to A/B test next month."
Get this

5 hook-tested variations grounded in your actual winner. Heist tracks which ones perform.

Edit + republish

Try something like
"Find my pricing blog post, add an FAQ section at the bottom answering common pricing questions, and republish to WordPress."
Get this

Edits land in WordPress without you opening the WP editor.

Try something like
"Find all my draft blog posts older than 30 days and add a CTA paragraph to each promoting our newsletter signup."
Get this

Preview the batch before it commits — and if you change your mind after, ask the agent to undo it.

Try something like
"Pull up my 'How to Use Heist' blog. Update the tool count to 85 and republish."
Get this

Specific, surgical edits with a single instruction.

Brand brain + voice

Try something like
"Read my last 10 posts and tell me where my voice drifts. Are there any 'AI-flavored' sentences I should rewrite?"
Get this

A drift report citing exact lines. Heist scores every post against your brand voice.

Try something like
"Remember this for future content: my company won the 2026 SaaS Innovation Award in March. Add it as a memory fact."
Get this

Fact persists forever. Every future draft can reference it.

Try something like
"Add a rule to my Brain: never use the word 'leverage' in my posts. It sounds corporate."
Get this

Server-side guardrail. Heist refuses to ship a draft that violates it.

Walk back mistakes

Try something like
"I just bulk-deleted 14 drafts I shouldn't have. Find the action and revert it."
Get this

Heist's audit log makes most writes reversible. Just ask the agent to undo it — drafts come back.

Try something like
"Yesterday I edited my pricing blog post and I want to roll it back to the version before that edit."
Get this

The agent finds the agent_action and restores the before-snapshot.

Try something like
"I added a memory fact about a discount code that's no longer valid. Find it and remove it."
Get this

Brain stays accurate over time without manual cleanup sessions.

Best practices

Tips for talking to Heist.

Short habits that make the agent useful instead of frustrating. Most are obvious in hindsight; not all are obvious up front.

Tip 01
Start with "what can you do?"

On a fresh chat, this prompts the agent to lay out everything it can do for you — what tools it has, what permissions, where to start. 30-second orientation that saves a lot of trial and error.

Tip 02
Be specific about platforms and counts.

"Plan a 2-week LinkedIn campaign — 3 posts per week" beats "make me some content." The agent uses Heist's per-platform rules (char limits, hashtag conventions, best posting times) — give it constraints and it does the rest.

Tip 03
Reference your top performer.

"Match the voice of my best LinkedIn post from the last 60 days" pulls your actual high performer and grounds new drafts in its tone. Better than "be casual" — it's your real voice, scored by your real audience.

Tip 04
Use approval mode wisely.

Brand-Brain writes (facts, rules, document uploads) queue for approval by default — they shape every future draft, so a human gate is sensible. Flip Settings → AI → Brain-write approval to instant when you trust the agent. Drafts go straight to your library either way.

Tip 05
Reserve "publish" scope for trusted automation.

When you connect, the OAuth flow grants read, draft, schedule. Drafts land in your library, you review, you publish. Only grant publish when you're comfortable letting the agent push to Meta / X / LinkedIn / WordPress without you in the loop.

Tip 06
When in doubt, dryRun first.

Bulk operations support a preview mode. Ask the agent to "preview the batch first" before it actually writes anything — it validates each item (brand access, content limits, your rules) and reports back without committing. You decide whether to ship it.

Tip 07
Mistakes are reversible.

Every action Heist takes is logged. If you said "delete those 14 drafts" and regretted it, ask the agent to revert it — Heist brings them back. Most actions are reversible (creates, deletes, updates, fact writes). Publishes already hit the public internet, so for those, ask the agent to republish with the correction.

Frequently asked

The honest answers.

What AI tools does Heist work with?
Heist works with any AI assistant that supports MCP. Today that's Claude (desktop and web), ChatGPT, Cursor, Copilot, and Windsurf. More clients support MCP every month.
Do I still need the web app?
No. Everything you can do in the Heist web app — plan, draft, schedule, publish, edit your Brain — you can do from your AI assistant. The web app is there when you want it, not because you need it.
Can the AI mess up my brand?
No. Every AI action lands in an audit log. You can set a preference so nothing applies until you approve it. All writes require a source or reason — the AI can't save a fact without citing where it came from.
What's the learning curve?
If you can have a conversation, you can use it. You chat naturally — ask anything you'd ask Claude, in your own words. We also publish named shortcut prompts (daily briefing, weekly plan, brand audit, voice refresh) that AI clients surface in their prompt picker, but they're optional — the agent figures out which Heist tools to call either way.
Do I have to use specific prompts?
No. The chat works exactly like talking to Claude. Ask anything, in any phrasing — the AI picks the right Heist tools automatically based on what you're asking for. The prompts on this page are examples, not requirements. Steal them, paraphrase them, or ignore them and just talk.
Is it safe to give an AI write access to my brand?
Yes — by design. When you connect via the OAuth URL flow, Heist's consent screen lets you pick exactly which brands the AI can access and what it can do (read brand context, create drafts, schedule, publish, edit your Brain). You grant only what you need; the AI can't escalate. Every write lands in an audit log and can be reversed by asking the agent to undo it.
Do I need an API key?
Not for Claude.ai or ChatGPT — those use the OAuth URL paste flow described above. API keys are only needed for stdio clients like Claude Desktop, Cursor, or Windsurf, which don't speak the web OAuth flow. The API-key path is hidden under Advanced Settings on purpose; most users never touch it.
What if the AI makes a mistake?
Most writes are reversible. Every action lands in your audit log; ask the AI to revert it and Heist restores the previous state. Brand-Brain writes (facts, rules) queue for approval by default, so they need your sign-off before they take effect.
Can I revoke access?
Yes. Heist → Settings → Connections → Connectors lists every active session. Click Revoke and the AI immediately loses access. Or revoke from inside Claude.ai / ChatGPT in their connector list.
What happens if I cancel?
Your data is yours. Your posts stay published on their platforms. Your Brain content can be exported. Cancel from Settings — no fees, no questions.
Ready?

Still running your content business from 5 tabs?

Set up your Heist Brain in 10 minutes. Connect it to Claude in 30 seconds. Run your content week in a 5-minute conversation tomorrow morning.

Start free
7-day free trial · No credit card · Cancel anytime