How to get things done with HexaHQ.

HexaHQ turns useful work done with AI into knowledge and workflows your organization can reuse. Your AI can save the facts, preferences, and repeatable steps that mattered as it works. You decide what stays personal and who can use the rest. When a task needs another system, HexaHQ can run approved capabilities under your organization's rules.

You use HexaHQ in two places. The AI is where you ask for work in plain language. The Hub is where you review knowledge and workflows, choose who can use them, manage connections and approval rules, and inspect activity. See the product overview.

HexaHQ has four building blocks. Knowledge is the context, connectors are the tools HexaHQ can reach, capabilities are the actions it can take in them, and workflows combine those into a job you can run again. Each one has its own page in Core concepts below, and the two minute building blocks walkthrough shows them in order.

You reach all of it through two places. The AI is where you ask, in plain words. The Hub is where you review it: you don't write the knowledge or build the workflows by hand, your AI does that as you work, and the Hub is where you see what it made, approve or adjust it, and share it with your team.

HexaHQ speaks the Model Context Protocol. Use HexaHQ through ChatGPT, Codex, Claude Chat, Claude Code, or Claude Cowork today.

The easiest path: create an account and HexaHQ hands you a one-click connect link for the client you use. To add it by hand, point your client at the endpoint https://mcp.hexahq.ai/mcp/. Each client adds an MCP server its own way, so follow that client's instructions for where the endpoint goes. Either way it uses OAuth, so you approve access in the browser the first time your client connects.

Step by step for Claude on your own plan or on Team or Enterprise, and for ChatGPT. Full walkthrough on How it works.

A settings change you make yourself, under Customize. Connecting happens in the browser, so nothing secret is pasted into Claude.

Adding HexaHQ to Claude on your own plan, with a walkthrough video and every screen shown.

Two jobs, not one. Only an Owner can add a custom connector, and they do it in Organization settings for the whole workspace. Each member then connects their own account, which is the half people miss: the connector being in the list does not mean you are connected to it.

Adding HexaHQ to Claude Team or Enterprise, with a walkthrough video and both halves written out.

Two settings screens in two places. Developer mode goes on under Security and login, which is not where plugins live, and then the plugin is created from the Plugins page. Create stays greyed out until you tick the acknowledgement, which reads like a rejected address.

Adding HexaHQ to ChatGPT, with a walkthrough video and every screen shown.

Start free: connect an app and get your first task done, no credit card. Already have an account? Sign in. See what's in Free vs Enterprise on pricing.

For tools with a standard integration, connecting is one click: ask your AI to connect the tool (for example, "connect Salesforce"), and HexaHQ gives you a secure browser link where you sign in and grant access. The credential is stored server-side. Nothing is pasted into the chat.

See the tools HexaHQ connects on Connect anything, and what a connector is on Connectors.

Some tools authenticate with an API key instead of OAuth. HexaHQ hands you a secure browser link where you enter the key. It is stored server-side and never travels through the chat.

Never paste a key, token, or secret into the conversation. By design there is no way to hand a raw key to your AI; credentials are always entered out-of-band in the browser. More on how connectors handle credentials and on Security & governance.

Most tools don't ship an MCP server. That's HexaHQ's edge. Describe the tool's API and your AI authors the capability to call it. You build and host no server, and write no code. Tools that already ship an MCP server connect directly.

How this works: Connect anything.

When you register your own OAuth app, the tool's developer console asks where HexaHQ lives: a host domain, a launch URL, a disconnect URL, and a connect or reconnect URL. Intuit (QuickBooks) needs all four before it releases production credentials. Google and Microsoft ask for a home page and a privacy policy.

The values are on HexaHQ's app URLs for an OAuth developer console, along with where each one goes per console. Two of them are not what you would guess: the disconnect URL names the tool and the connect URL does not, and the redirect URI is different for every company, so read that one off your own setup page rather than copying it.

Credentials are entered in the browser and held server-side. No key transits the chat, and there's nothing to commit to git. Access is keyed per person: a connector is shared once, but each teammate connects with their own credential and acts as themselves.

Details on Connectors and Security & governance.

Just ask in plain language: "prep me for today's customer meetings," "draft follow-ups for yesterday's calls." HexaHQ figures out which tools and steps the task needs, runs the ones that are auto-approved, and pauses for your approval on anything that writes, sends, or deletes.

For 400 real examples by team, see use cases.

Do it once in a conversation. HexaHQ captures the multi-step work as a workflow (in your voice and format), so next time it's a single sentence, for you and anyone you share it with. The workflow lives in your organization, not one person's chat history, so it doesn't evaporate when someone leaves.

What a saved process contains, and how it runs: Workflows.

Share the workflow and your teammate runs it on their own credential. They never see yours. One technical person can connect the tools and author the capabilities once; everyone else clicks a link, signs in with their own account, and runs the workflows. No JSON, no setup.

More on the team story: HexaHQ for teams.

• A capability is one action in a tool: look up a record, send a message.
• A workflow is a saved process built from capabilities and run with one sentence.
• Knowledge is the context both of them draw on, so results come back in your voice and format.

This is the one idea worth understanding properly, because everything in HexaHQ works this way: knowledge, workflows, and capabilities all follow the same path. Who can see a document is a separate question, answered by collections.

There are two layers.

• Your copy. An edit you make to something that already exists is yours. It applies to your work right away, with nothing to approve. Nobody else sees it, and nobody else's results change.
• The company copy. What everyone gets by default.

When you run something, HexaHQ uses your copy if you have one, and the company copy if you don't. So you can adjust anything for yourself without asking, and without any risk of changing your teammates' work.

New knowledge is offered to the company as it is written, because knowledge is the company's shared memory and a document nobody else can read helps nobody else. On a company that has not turned review on, it is available to everyone straight away. Anything marked Personal only stays with you, and you can turn automatic sharing off for yourself in your personal settings. Full detail on Knowledge.

Workflows are the other way round. A workflow acts on your systems, so publishing one to the company is always somebody's decision and is never automatic. The same is true of a company that switches knowledge review on: sharing then puts the document in a queue for an admin.

If the company copy changes while you have your own version, your edits are not thrown away. HexaHQ merges the new company version underneath your changes and gives you an updated copy. If the two genuinely conflict, it comes back to your AI to sort out with you. Nobody's work gets quietly overwritten.

One thing that surprises people. A knowledge doc can be attached to a connector or to a group of related actions, so it applies to all work in that area, including work you never attached it to yourself. That attachment is shared across the company on purpose: it is how a good writeup starts helping everyone instead of only the person who wrote it. It only reaches your teammates once the doc itself is a company doc. If it is still just your copy, it guides your work and nobody else sees it.

Admins can see all of it in the Hub: what is waiting for review, what the company copy currently says, and the history of what changed. If something started applying that shouldn't have, that is where to look.

If your engineers use a coding agent like Claude Code or Codex, they already keep instructions in files in the repo. HexaHQ is a second home for context, so it helps to know which goes where.

The rule: put it where it stays true, and where the people who need it can reach it.

• Files in the repo for anything the code next to it proves: architecture, code conventions, API contracts, runbooks, why a module is built the way it is. It travels with the code, gets reviewed in the same pull request, and is wrong the moment the code changes, which is exactly the signal you want.
• HexaHQ for anything the whole company needs: product specs, positioning, pricing, brand and tone, sales and support playbooks, what you have promised customers, how a process runs. Sales, marketing, and support never open the repo, and knowledge in HexaHQ reaches every AI tool in the company, not only the coding agent.

A few worked examples. An endpoint's contract goes in the repo; what the feature is for and who it is sold to goes in HexaHQ. A migration runbook goes in the repo; the promise you made customers about downtime goes in HexaHQ. Error-handling conventions go in the repo; the tone of the error message a customer reads goes in HexaHQ.

This is a boundary, not a split. Code that needs product context still gets it: a coding agent connected to HexaHQ can read the spec, the positioning, or the support policy while it works. So keeping specs in HexaHQ costs your engineers nothing. The repo keeps the how, HexaHQ keeps the what and the why, and the agent bridges them when it needs to.

This is also why it is worth capturing something the moment you learn it. Your AI can write it to HexaHQ as it works, and what it writes starts as your own copy and only reaches the company after review, the same as everything else. See your copy vs the company copy.

Every capability runs under an org-wide policy. Allow runs automatically (typically read-only lookups). Ask pauses for your approval first (anything that writes, sends, or deletes). Deny blocks it outright. The gate carries each person's own identity, so approvals and actions are always attributable.

The policy belongs to the capability, so it applies wherever that action is used. See Capabilities, or Security & governance.

When a task hits an "ask" step, your AI shows you exactly what it's about to do and waits. Approve it and HexaHQ runs the step; decline and it stops there. Nothing is sent or changed without your say-so.

Open Connectors in the Hub, select the tool, and use the Manage menu to clear your saved sign-in. HexaHQ can no longer act in that tool as you. It affects your own sign-in only: teammates keep theirs, and the tool stays set up for your organization.

Ending the grant on the other side is a separate step, done in that tool's own account settings. See Disconnect, revoke, delete for what each one does.

HexaHQ holds three separate things: your saved sign-ins, a record of what ran, and the knowledge and workflows built up as you work. You can remove any of them without touching the others.

Managing and deleting your data walks through each one, and Privacy request is how you ask us to delete the rest.

Every action HexaHQ takes is recorded on a run timeline (what ran, who asked, and when), so there's always an answer to "who did what." You can see your own work; admins read the whole organization's history in the Hub, under Activity Log. See Runs for what a run holds, or HexaHQ for teams.

The Hub is HexaHQ's admin web app. It's where you review and approve proposed workflows and connectors, manage knowledge, set the approval policy, and read the audit timeline, the place to review and govern the work, alongside the AI where it happens. See HexaHQ for teams.

As people use HexaHQ, it proposes new workflows and connectors. Admins review them in the Hub and approve the ones the org should share, so good process spreads deliberately rather than by accident.

A connector is shared once but keyed per person: each teammate connects with their own credential, so actions carry their own identity and permissions. Nobody borrows anyone else's access. This is what Claude Teams can't do today. See HexaHQ vs Claude Teams.

Free is for individuals: connect your tools and run workflows. Enterprise adds org-wide governance and the Hub. See pricing; Enterprise is a conversation with our team (no self-serve flow).

Your data is processed only to run the tasks you invoke, and is never used to train models. For exactly what's stored, how long, and which third parties are involved, see the Privacy Policy and the subprocessors list. More Q&A on the FAQ.

Can't find what you need? Contact us or email sales@hexahq.ai and we'll help.

First check the tool is connected: ask your AI to connect it, or connect it from the Hub. If the tool is connected but the specific action doesn't exist yet, ask your AI to do it anyway: it can author the capability from the tool's API. If a step is blocked, it's likely a deny policy. An admin can adjust it in the Hub.

That's the approval policy doing its job. "Ask" actions wait for your approval; "deny" actions are blocked org-wide. An admin sets and adjusts the policy in the Hub. Open the run to see which step stopped and why, or read how approvals work.

Credentials can expire or be revoked at the source. Reconnect the tool the same way you first connected it (ask your AI to connect it, or use the connect link from the Hub), then re-authorize in the browser. Your workflows keep working once the credential is refreshed.

This is about a tool HexaHQ connects to, like your email or your CRM. If the problem is that your AI is working from an out of date list of what HexaHQ can do, that is a different fix: refresh the HexaHQ connector in Claude or ChatGPT.

Claude re-reads HexaHQ's tool list every time it connects, so a change on HexaHQ's side is normally picked up by starting a new chat, with nothing to click. When that has not worked, there is a Refresh tools list control in the connector's own settings.

Refreshing the connector in Claude, with a walkthrough video and the steps written out.

ChatGPT holds a cached copy of what HexaHQ can do, so refreshing takes two steps that are easy to miss: a Refresh button at the very bottom of the plugin panel, and then a new chat, because a chat you already have open keeps the copy it started with. The Reconnect item in the menu at the top looks like the right control and is not.

Refreshing the connector in ChatGPT, with a walkthrough video and the steps written out.