Working in HG MCP Server — end user
0. Who is this documentation for?
What this doc covers
You are in and want to start using HG MCP Server. This page shows you what you can do in the app itself: asking questions, running agents, and seeing what your organization is using.
Prerequisites
You already have an account. You registered with your corporate email and have access to the MCP tools.
1. How to get to the HG MCP Server app
Go to phoenix.hginsights.com and click App in the header. Sign in with your work email address, or with Google or Microsoft. You land on your organization's home page.
What the app is, next to using the tools from your LLM
Most of the time you will use HG MCP Server from inside an LLM, with the tools connected to it. The app is the place around that.
Asking questions is the same in both. The Playground in the app is the same conversation with the same tools behind it, so anything you can ask there you can ask from your LLM, and the other way round — when your LLM is connected with an ordinary user key. (An admin-scoped connection is a separate, narrower surface that carries only the admin tools; the Playground always runs at user scope, so the two do not overlap.)
What the app adds is everything around the asking. It is where your work is kept and where you see what is going on: saved prompts and past conversations, agents you can run and the files they produce, your own API key and the endpoint your LLM needs, and what your organization is consuming. Think of it less as a second way to ask a question and more as the console for your account.
2. What you can do
Below are the tabs displayed in the sidebar as a member of your organization.
| Tab | What it is | Where to read |
|---|---|---|
| Home | Where you land. The box asks what you want to know; your question is carried straight into the chat, so you do not type it twice. | See 3 |
| Playground | A chat window with your organization's tools behind it. | See 3 |
| OttoBot | Your organization's own assistant. Only appears if your organization has OttoBot enabled — not everyone sees this. | See 4 |
| Agents | Saved work that runs on its own, and the files it produces. | See 5 |
| Subscriptions | Sending signals out to another system. | Not covered here, your admin |
| Integrations | The data sources the tools draw on. | Not covered here, your admin |
| MCP | Your own API key, and the endpoint address your tools need. | See 6 |
| Settings → Teams | The teams in your organization and who is in them. | Not covered here, your admin |
| Settings → Usage | Credits used, the limit, and what is left. | See 7 |
| Settings → Onboarding | Approving colleagues who asked to join. | Not covered here, your admin |
| Settings → Branding | How the organization looks. | Not covered here, your admin |
| Settings → OttoBot | The assistant's configuration. Only appears if your organization has OttoBot enabled. Most of it is admin-only, but Analytics and Sessions are open to you — Sessions shows past conversations across your organization, including colleagues'. | See 4 |
3. Playground
Click Playground in the sidebar. The Playground is also what the Home page directs you to.
This is a chat window with your organization's HG MCP Server tools behind it: ask in plain language and it decides which tools to call and shows you what came back. This is equivalent to using HG MCP Server within an LLM like Claude or Codex.
A good first try is to ask for firmographic data on a company you know, your own, or a customer, and watch which tools it reaches for. For example: "Give me the firmographics for HG Insights." The suggestions on the home page are there for the same reason.
What you can do here
| You can… | How |
|---|---|
| Keep going in the same conversation | Ask a follow-up and it keeps the context. You do not have to restate the company or the question. |
| Come back to an earlier conversation | A Conversations panel, opened from the header, groups past chats as Today, Yesterday, This Week and by month. Nothing is lost when you close the tab. |
| Save a prompt you use often | Saved Prompts keeps your own wording so you can re-run it without retyping. You can search them and delete the ones you no longer want. |
| Limit which tools it may use | MCP Tools lets you narrow the set. Useful when you want a specific source rather than whatever it judges best. |
| Choose which model answers | A model picker sits at the top. A default is chosen for you that suits a first pass; change it if you have a reason to. |
| See what it actually did | Each answer shows which tools ran, as they run. This is how you check where a number came from. |
If a tool needs an integration your organization has not connected, the tool returns an error naming what is missing rather than failing silently. (A banner about missing connectors appears when you open the Playground from a workflow link, not on every question.) Connecting an integration is open to any member, so this may be something you can resolve yourself — see Integrations.
What you can do in the Playground is what you can do anywhere else. It runs as you, with your own permissions, it is not a way to reach data you would not otherwise see.
Phoenix credits are consumed by the tool calls, not by the conversation. Credits are spent on the data each tool call retrieves, exactly as they would be from any other client — the assistant reading your question and deciding what to call consumes no Phoenix credits, however long the exchange runs. It is not free in absolute terms: the model itself runs on your organization's own configured provider credential and is billed by that provider.
4. OttoBot
Some organizations have OttoBot, a chat assistant of their own. If yours does, it appears in the sidebar just under Playground, and the question box on the home page sends your question there instead.
It works like the Playground, with one difference: your organization has given it standing instructions of its own, so it answers in the way your company has decided it should. You use it the same way, by asking.
If you do not see it, your organization does not have it turned on yet. Contact your HG representative to enable it.
5. Agents
Click Agents in the sidebar. An agent is a saved piece of work that runs on its own rather than in a back-and-forth chat.
Before you run one you can read what it does: the Library filters by category and by use case, and each agent shows its details — plus, where the agent defines them, the tools it is allowed to use, the model it prefers, and sample output. Run it, and the result lands in Artifacts.
| Section | What it is for |
|---|---|
| Library | The agents available to your organization. Start here. |
| Builder | Create or change an agent. Open to any member. |
| Artifacts | The files an agent produced. Open or download them from here. |
| Logs | What each run did, and how it ended. The first place to look when a run did not produce what you expected. |
| Analytics | How much the agents are being used. |
Building an agent is open to any member of the organization — there is no admin restriction. The Builder tab is greyed out until your organization has at least one agent, unless you are already in the Builder view: fork a blueprint from the Library first, and the tab becomes active.
Deleting an artifact is the one action here reserved for admins. The run itself is kept; only the files go, and they do not come back.
6. Your key and the connection details
Click MCP in the sidebar. This is where your own API key lives, alongside the endpoint address your tools need.
| You can… | How |
|---|---|
| Create a key for yourself | Give it a name and create it. You do not need to be an admin. Copy it when it appears, and if you need it again it stays visible in the key list on the same page. |
| Copy the endpoint address | It is on the same page, beside the keys. |
| See when a key was last used | The list shows each key's name, when it was created, and when it was last used. |
| Revoke a key | Revoke it from the same list. Do this if a key has been shared or you no longer need it. |
Keys come in two scopes. User is the default and the one you want. Admin is selectable only if you are an org admin — the button is greyed out otherwise, and a request that reaches the server anyway is refused with "Only org admins can mint admin-scoped API keys." A non-admin cannot mint one. (A key minted while you were an admin is not revoked if you are later demoted — it simply stops behaving as an admin key.)
7. What your organization is using
Click Settings → Usage. This shows the whole organization, not just you.
| What you see | What it tells you |
|---|---|
| Current usage | Credits used, the credit limit, and what is left. This is a running total since HG last started a new period for you, not a figure that resets on a date of its own. |
| Daily usage | The last 30 days, day by day. This is where a spike shows up. |
| Usage by tool | Which tools the credits went to. |
| Usage by authentication source | Whether calls arrived on an API key or through an OAuth connection, plus an unattributed bucket visible under the "All" filter. This splits by credential type, not by app-versus-client — the Playground's own calls are attributed to whichever credential your account uses. |
| Cache performance | The hit rate and the credits it saved you. Cached answers cost nothing. Shown alongside it is Avg Response Time, how long calls are taking. |
The credit limit is shared, not a personal quota. If your organization is on hard enforcement, a tool call is refused once its cost would take the organization past the limit — so the expensive ones stop first, for everyone, while the balance still shows credits left — until HG raises the limit. It does not clear on its own. Free lookups and answers already in the cache keep working, so the app will not feel completely dead. On soft enforcement nothing is blocked, but overages may be charged. Either way, if you can see the limit getting close, tell your admin rather than waiting to find out which applies to you.
8. If something goes wrong
Getting in is covered separately. This is what you may meet once you are using the app.
| What you see | What it means | What to do |
|---|---|---|
| A tool errors saying an integration is missing | The data source behind that tool is not connected for your organization. | Open Integrations and connect it — this is open to any member, not just admins. If you are unsure which credential to use, ask your admin. |
| A tool you expected is not offered at all | Either its data source is not connected, or your organization is on an older set of tools. | Tell your admin which tool you were looking for. They can tell the two apart. |
| The tools that cost credits stop, for everyone at once | The organization has reached its credit limit and is on hard enforcement. The costly tools stop first, so this can begin while the balance still shows credits left. Free lookups and cached answers still work, so only part of the app goes quiet. | Retrying will not help, and it does not clear on its own — your admin has to ask HG to raise the limit. Check Settings → Usage to confirm that is what happened. |
| "Sending too fast — wait a moment and try again." | A per-person limit on how quickly you can send messages. | Pause for a moment. It clears on its own, and there is nothing to raise. |
| "Your session expired — please refresh the page." | You were signed out. | Refresh and sign in again. Nothing is lost, your conversations are kept. |
| "You don't have access to this organization's OttoBot." | You are signed in, but the address you opened belongs to an organization you are not in. | Check you are on your own organization's page. If you should have access to that one, ask its admin to add you. |
| OttoBot is temporarily unavailable | The service is not responding. | Try again in a few minutes. If colleagues have the same problem, report it once rather than everyone retrying. |
| An agent run did not produce what you expected | The run may have failed partway, or a tool it needed was unavailable. | Open Logs and read that run. It shows what it did and how it ended. |
| A link to a file has stopped working | Not expiry — artifacts are kept indefinitely. The run or its files may have been deleted by an admin. | Ask an admin, or re-run the agent. |
| You were sent to a setup page the first time you opened the Playground | Your account had no key yet. | Follow the page once. Your question is carried through afterwards, and it does not happen again. |
For anything this page does not cover, ask an admin in your organization first, they can see more than you can. If they cannot resolve it, they will bring in HG.