Model Context Protocol is an open standard that lets an assistant use tools from other services. AnswerGravity offers its own: your assistant calls them to read your brand's data when you ask.

What do I need to connect it?

A client workspace with a platform plan and an admin who connects. In Claude your account is enough (OAuth), no key. In other clients, an API key generated in Account settings → API & MCP; it is shown once and generating another replaces it.

Claude on the web, desktop and mobile, with your account and no key (OAuth): disconnect it whenever you like in “API & MCP”, under “Connected applications”. And Cursor, Claude Code, Codex, VS Code and any MCP client that uses Streamable HTTP with a key.

Can it change anything in my account?

By default it only reads. If you check ‘Allow question management through MCP’ on the key, or allow it when connecting Claude, it can also add questions or stop tracking them, always with a review first and keeping the history. It never publishes, contacts anyone, changes your plan or spends credits.

Does it cost anything?

Queries spend no credits. Questions you add use your plan's question quota and measurement budget, as they do from the console.

What data does my assistant see?

Only what the tools it calls return, within the permissions of the connection and your role: questions, saved answers, sources, competitors, listening, drafts, orders and reports. What the assistant then does with that data depends on its provider.

MCP connection

Your brand data, in your assistant

Read questions, answers, sources, competitors and social listening from Claude, with your account, or from any MCP client, with an API key from your workspace. You can also authorize question management, with prior review.

Open API and MCP in your workspace See the configuration

How to connect

With your account or with a key

You need an AnswerGravity client workspace with a platform plan; if you do not have one yet, see the plans. A workspace admin connects, and every tool respects the permissions of the connection and your role.

Claude
With your account, no key
Claude on the web, desktop or mobile: add a custom connector with the MCP address and choose “Allow” in AnswerGravity. You sign in with your account, no key; only an admin connects, and you disconnect it in “API & MCP”, under “Connected applications”. When connecting you choose whether it may also manage questions.
Other clients
With an API key
Cursor, Claude Code, Codex, VS Code or any MCP client with Streamable HTTP: generate a key in “Account settings” → “API & MCP” and paste it into the configuration. It is shown once and replaces the previous one. It is read-only unless you check ‘Allow question management through MCP’.
Then
Start with list_brands
Use the brandId it returns to read questions, mentions, original answers and sources.

Claude · Settings → Connectors

1. Add custom connector
2. Name: AnswerGravity
   URL: https://answergravity.com/api/v1/mcp
3. Connect: sign in to AnswerGravity and choose “Allow”

~/.cursor/mcp.json

{
  "mcpServers": {
    "answergravity": {
      "url": "https://answergravity.com/api/v1/mcp",
      "headers": {
        "Authorization": "Bearer YOUR_KEY"
      }
    }
  }
}

In your terminal

claude mcp add --transport http answergravity https://answergravity.com/api/v1/mcp --header "Authorization: Bearer YOUR_KEY"

~/.codex/config.toml

# First, in your terminal: export ANSWERGRAVITY_API_KEY="YOUR_KEY"
[mcp_servers.answergravity]
url = "https://answergravity.com/api/v1/mcp"
bearer_token_env_var = "ANSWERGRAVITY_API_KEY"

.vscode/mcp.json

{
  "servers": {
    "answergravity": {
      "type": "http",
      "url": "https://answergravity.com/api/v1/mcp",
      "headers": {
        "Authorization": "Bearer ${input:answergravity-key}"
      }
    }
  },
  "inputs": [
    {
      "type": "promptString",
      "id": "answergravity-key",
      "description": "AnswerGravity API key",
      "password": true
    }
  ]
}

mcp.json

{
  "mcpServers": {
    "answergravity": {
      "type": "streamable-http",
      "url": "https://answergravity.com/api/v1/mcp",
      "headers": {
        "Authorization": "Bearer YOUR_KEY"
      }
    }
  }
}

The address is also shown next to your key, under Account settings → API & MCP.

Check the connection, and withdraw it whenever you like

Ask your assistant “List my AnswerGravity brands”. If it answers with your brands, the connection works: it used list_brands.
If it says it has no access, the Claude connection was disconnected or the key no longer works because a new one was generated: connect again or paste the new key.
To withdraw Claude's access, choose “Disconnect” under API & MCP → Connected applications. To withdraw a key, generate a new one: the old one stops working at once.

What to ask

Ask your assistant

Once connected, ask in your own words for what you would see in the console. The assistant picks the tools and answers with your brand's data.

“Which questions does ChatGPT not name me in this month, and which brands appear instead?”
“Which domains do the answers about my category cite most, and which of them name my competitors?”
“Compare my visibility with my two main rivals over the last 30 days.”
“Show me Gemini's original answer to my main question, with its sources.”
“What new conversations are there about my keywords, and on which forums?”
“Suggest three questions about pricing and add them once I have reviewed them.”

Scope

What it can and cannot do

It can

Read questions, mentions, original answers and sources
Read competitors, keywords, drafts, orders and reports
Check your workspace limits and monthly draft allowance
Use up to 35 read tools, depending on the connection’s permissions, with pagination on lists
With your authorization, review and add questions (preview_add_prompts and add_prompts) or stop tracking (preview_stop_prompt and stop_prompt), keeping their history. These are 4 additional tools, up to 39 in total, depending on your permissions; new questions use your plan’s capacity and measurement budget

It cannot

Publish content or contact anyone
Change your plan or consume credits
Bypass the permissions of the connection and your role
Create tasks or edit drafts

See the roadmap

Reference

All 39 tools

35 read tools and 4 to manage questions, grouped like the console's screens. Each one respects your key's and your role's permissions.

Brands and account

Your brands, plan, usage and team.

list_brandsRead
Lists the workspace's active brands and their measured visibility.
get_brandRead
Reads an active brand and its measured visibility.
get_subscriptionRead
Reads the assigned plan and its limits. A saved price does not activate a subscription.
get_balanceRead
Reads allocated and used service credits. They are not withdrawable money.
get_usage_metricsRead
Reads usage and actual limits for questions, keywords, members and models.
list_team_membersRead
Lists members and invitations, with their emails. Requires the members permission.

Visibility and questions

How you appear, how it changes and where to act.

get_visibility_overviewRead
Reads the selected brand and its share of voice from valid stored measurements.
get_visibility_trendRead
Reads the 7, 30 or 90-day visibility series by model that the console uses.
list_promptsRead
Lists tracked questions, their models and results. “Not measured” means there is no valid evidence yet.
list_opportunitiesRead
Reads the same opportunities as the console (cited sources, open conversations and measured questions), ordered by impact. It never searches, measures, drafts or spends credits.
list_prompt_responsesRead
Reads the console's answer history with filters and pagination. It reads stored evidence only and never starts a measurement.
get_prompt_gapRead
Reads «Why them and not you» for one question: the answers that do not name you, who they name and which pages they cited, with one next step per kind. It reads stored data only.
get_brand_comparisonRead
Reads the comparison of the brand with up to three rivals from closed reports, over 1 to 90 days. Choosing rivals does not start tracking them.

Answers and sources

The original answers, their provenance and the sites they cite.

list_llm_mentionsRead
Lists the stored observations for a tracked question, with status and run time.
get_llm_responseRead
Reads a stored answer and its provenance. Raw provider payloads are excluded; a long answer says when it was truncated.
list_llm_citationsRead
Lists the domains cited in stored answers: how many answers cite each one and how many of them name your brand. That is co-appearance in the answer; the cited page is not read.
list_llm_modelsRead
Reads the models configured for an active brand.

Competitors

Your competitors and their share of voice.

list_tracked_competitorsRead
Lists tracked competitors and their measured comparison metrics.
get_competitor_leaderboardRead
Reads competitors' share of voice from stored measurements.

Social listening

Your keywords and the threads found.

list_tracked_keywordsRead
Lists active social listening keywords.
list_mentionsRead
Lists stored listening threads, as in the console. It does not run an external search.

Drafts and tasks

What has been prepared and its status. Nothing is published from the MCP.

list_draft_ordersRead
Reads the brand's saved AI drafts, twenty per page. It does not generate or use allowance.
get_draft_orderRead
Reads one draft and its delivered versions. Delivered means saved text, never a published post.
get_draft_allowanceRead
Reads the plan's monthly draft allowance and its renewal. It does not reserve allowance or spend credits.
list_tasksRead
Reads the legacy community queue. Reading does not publish or allocate credits.
get_taskRead
Reads a draft or an action from the legacy queue.
list_platform_tasksRead
Reads platform task tracking: open drafts and orders with their status.
get_platform_taskRead
Reads a platform order, its history and edits. A simulated order is not a public delivery.
get_task_draftRead
Reads a saved platform draft. It does not save or submit it.
list_task_groupsRead
Reads platform task groups and their orders. It does not submit or run them.
get_action_ledgerRead
Reads «What you've done»: the brand's actions and what AI answers showed afterwards. A cited page is a fact; a before and after is what was observed, never a cause. It adds or deletes nothing.
get_task_groupRead
Reads a saved group and the status of its orders.

Media and reports

Verified editorial inventory, orders and saved reports.

list_backlink_listingsRead
Lists verified editorial inventory. An empty result means no verified inventory is available.
list_backlink_ordersRead
Lists editorial orders and their current status.
list_reportsRead
Lists a brand's saved reports. Reading does not create a public link.

Question management, with your permission

Only if the key allows it. Every change goes through a preview and your explicit confirmation.

preview_add_promptsWrite
Reviews the questions, duplicates, selected models and remaining plan capacity before adding them. It does not start tracking or measuring.
add_promptsWrite
Adds the reviewed questions with the preview key and your confirmation. It rechecks permissions, models and capacity; they join the plan's scheduled measurements, with no purchase or immediate call.
preview_stop_promptWrite
Reviews the active question before stopping its tracking. Existing evidence is kept.
stop_promptWrite
Stops future tracking of the reviewed question with the preview key and your confirmation. It does not delete its history or change the plan.

FAQ

Common questions

About the connection, permissions, cost and data.

Can't find your question? Write to us

Model Context Protocol is an open standard that lets an assistant use tools from other services. AnswerGravity offers its own: your assistant calls them to read your brand's data when you ask.
A client workspace with a platform plan and an admin who connects. In Claude your account is enough (OAuth), no key. In other clients, an API key generated in Account settings → API & MCP; it is shown once and generating another replaces it.
Claude on the web, desktop and mobile, with your account and no key (OAuth): disconnect it whenever you like in “API & MCP”, under “Connected applications”. And Cursor, Claude Code, Codex, VS Code and any MCP client that uses Streamable HTTP with a key.
By default it only reads. If you check ‘Allow question management through MCP’ on the key, or allow it when connecting Claude, it can also add questions or stop tracking them, always with a review first and keeping the history. It never publishes, contacts anyone, changes your plan or spends credits.
Queries spend no credits. Questions you add use your plan's question quota and measurement budget, as they do from the console.
Only what the tools it calls return, within the permissions of the connection and your role: questions, saved answers, sources, competitors, listening, drafts, orders and reports. What the assistant then does with that data depends on its provider.

Your next step starts here

Measurements in ChatGPT, Gemini, Copilot, Perplexity and Google (plus Claude), evidence-backed priorities and included drafts to review. Choose your capacity and get started.

See plans

Start with a free check of your brand