Overview#
The MCP server, the API, and the command-line tool do what you do in the dashboard: read your campaigns and results, and start the same work.
- MCP server: connect Claude, ChatGPT, or another AI assistant.
- API: call Point11 from your own code at
https://point11.ai/api/v1. - CLI: use the API from your terminal.
Sign in to open the API reference; it opens for members of a Point11 organization. It lists every request, what it takes, and what it returns, and which dashboard actions the MCP server, API, and CLI can reach. To test ads and other creative against your customers' digital twins, see Test creative with the API. For coding agents that work the tickets Factory files from Advisor's actions, see Factory.
Connect an AI assistant#
Use this MCP server URL: https://point11.ai/api/mcp. Sign in to Point11 and approve access when prompted. Point11 then asks which accounts this connection works in. Your organizations and your Manager accounts are in one list: tick the ones you want. If you have one account, it is ticked for you. With one organization ticked, your assistant works there; with more, tell it which one to use. In Settings > Connections, under MCP, Connect beside Claude opens Claude's Add custom connector page in a new tab with Point11's name and this URL filled in, and Set up beside ChatGPT shows this URL and the steps. Any other app that works with MCP, such as Cursor, connects with the same URL and sign-in, and appears there by its name once it has. To change a connection later, open its Details there. To connect with an API key instead, use the steps for your client below.
Your assistant sees a few tools that reach everything you can do: find_tools searches Point11's tools for what you ask, describe_tool says what one of them takes, run_read_tool runs a tool that only reads, and run_tool runs any tool, with the same approvals as the tool itself. discovery_dashboard_views_get shows a Discovery dashboard view, and when you sign in, list_accessible_organizations lists the organizations and Manager accounts your assistant can use. If your assistant asks you before it uses each tool, you can always allow run_read_tool, so reads go ahead while every change still asks through run_tool. An assistant that calls a tool by its name keeps working, even with an older version of a tool; find_tools finds only the current version of each.
Registries and clients can read the server's description at https://point11.ai/.well-known/mcp/server-card.json.
Claude on the web and desktop#
- Open Customize > Connectors, then + > Add custom connector.
- Name it Point11, enter
https://point11.ai/api/mcpexactly, and select Add, then Connect. - In a conversation, open + > Connectors and enable Point11.
Instead of steps 1 and 2, you can select Connect beside Claude in Settings > Connections: Claude's Add custom connector opens in a new tab with the name and URL filled in. Select Add, then Connect, and go on to step 3.
For Team and Enterprise plans, an owner first adds the connector in Organization settings > Connectors. See Claude's connector guide.
Claude Code#
Run in your terminal:
claude mcp add --transport http point11 https://point11.ai/api/mcpThen open Claude Code, run /mcp, and select Point11 to authenticate. To use an API key instead, add it as a header:
claude mcp add --transport http point11 https://point11.ai/api/mcp \
--header "Authorization: Bearer $POINT11_API_KEY"ChatGPT on the web#
- Open Settings > Security and login and enable Developer mode.
- Open Plugins > +. Enter Point11, a short description, and the server URL under Connection, then create the connection.
- Review the tools, start a new conversation, and select Point11 from the tools menu.
Availability depends on your account and workspace settings. See OpenAI's connection guide.
ChatGPT desktop and Codex#
In the desktop app, open Settings > MCP servers > Add server. Name it Point11, choose Streamable HTTP, and enter the server URL. Save, select Restart, then Authenticate.
For the Codex CLI, run:
codex mcp add point11 --url https://point11.ai/api/mcp
codex mcp login point11To use an API key instead, name the environment variable that holds it:
codex mcp add point11 --url https://point11.ai/api/mcp --bearer-token-env-var POINT11_API_KEYDesktop, CLI, and IDE clients share the configuration on your computer. See OpenAI's MCP guide.
See your dashboard in your assistant#
Ask for a Discovery dashboard view, such as Brand share, Brand win rate, or Competitors. An assistant that shows interactive apps draws the same card you see in your dashboard, with a link to open it in Point11. Other assistants get a short summary of the same numbers and the link.
Sign in or use an API key#
- Sign in as yourself. Point11 asks you to approve access, then your assistant or tool acts with your role in the accounts you selected. Use this for AI assistants and your own work.
- Use an API key for software that works without you, such as a script or a scheduled job. A key acts for one organization and can only use the access you give it.
A few things always need you. Your own Alpha conversations and projects, comments, and Research study drafts belong to a person, so an API key can't reach them. Pausing or deleting a campaign, inviting someone, or removing a member asks you to approve it first. You see the organization and exactly what will change, and your approval counts only for that request. Billing, API keys, and member roles change only in the dashboard.
When Alpha needs you before it uses a connected service, its answer carries a card that only you can answer, in Alpha in the dashboard: Always allow, Approve once, or Don't allow for the action Alpha is waiting to take. Your assistant can't answer it for you. If the service has changed or added actions since you accepted them, the card names each one, marked changed or new, and Alpha can't see or use them until you choose Confirm; if the card can't be confirmed, connect the service again.
Work you start from the MCP server, API, or CLI counts toward your plan the same way it does in the dashboard.
Choose what each assistant can do#
Every AI assistant or app that has acted for you in an organization is listed in that organization's Settings > Security, under Access in this organization. Choose what each may do there: Full, Read-only, or Blocked. Blocked stops the app in that organization alone and leaves its connection as it is. Owners and admins also choose what members' assistants may do in the whole organization in Settings > Security > Organization under AI clients (MCP): Allowed, Verified apps only, Read-only, or Off, and revoke a member's app under Members' connected apps. Verified apps only admits only apps Point11 can verify, such as Claude, ChatGPT, and Cursor; an app that registered itself is listed as Unverified app and is refused, and Read-only admits only verified apps too. A change applies from the next request, and API keys aren't affected.
Your connections are in Settings > Connections, under MCP, one row for each app you connected; an app appears there a few moments after you connect it. Each app has one connection: connecting it again, on another device too, updates that connection to the accounts you tick, and the app keeps working everywhere you use it. Its Details shows which accounts the connection works in. A connection works only in the accounts you chose for it, your organizations and your Manager accounts, and in each only while you belong to it; in an organization, its owners must also allow the app. It gains one only when you add it there, even an account you join later. Select Details, tick its accounts, then Save. The change applies from the next request, with nothing to connect again. When a change adds something and you use 2-step verification, Point11 may ask you to confirm it's you first. Select Remove in Details to stop the app everywhere you connected it, until you connect it again.
Your assistant can give up an account or remove its own connection, but it can't add one: when it needs an account its connection doesn't reach, it can give you a link to that connection in Settings. Apps you connected before connections were listed keep working as they were; connect one again to manage it there.
Create an API key#
Admins and owners can create keys.
- Open Settings > API keys and select Create key.
- Enter a Name, choose its Access for each product, when it Expires, and whom it Belongs to.
- Select Create, then copy the key. You see it only once.
Keep the key in a secret manager, never in your code. An organization can have up to 10 active keys. To stop a key, select Revoke; Rotate gives it a new secret. A key that belongs to Me is revoked once you leave the organization or can no longer manage keys; one that belongs to The organization keeps working.
Make your first request#
Send your key as a bearer token:
curl https://point11.ai/api/v1/discovery/campaigns \
-H "Authorization: Bearer $POINT11_API_KEY"Requests that read use GET. Requests that change something use POST with a JSON body and an Idempotency-Key header holding a new UUID:
curl https://point11.ai/api/v1/inspector/scans/start \
-H "Authorization: Bearer $POINT11_API_KEY" \
-H "Idempotency-Key: $(uuidgen)" \
-H "Content-Type: application/json" \
-d '{"url": "https://example.com"}'If you sign in instead of using a key, also send Point11-Organization with your organization ID. Copy it from Settings > API keys.
Handle responses#
- Retries. If a request is refused with a
4xx, fix what it names and send it again with the sameIdempotency-Key: it runs. If it fails with a5xx, such as a503, send it again with the same key: creating a webhook or a site, a start, an upload, or a send runs again safely; any other change runs nothing and answers409request_repeated, because Point11 can't tell whether the first attempt made its change. Once a change has succeeded, sending it again with the same key runs nothing and returns the first answer, so you still learn what was made if your connection dropped. Point11 keeps that answer for 30 days. Some requests answer with what they made instead: creating a webhook or a site answers with the item as it stands now, without any secret it showed the first time, and a start, an upload, or a send answers with the same run, upload, or send. Nothing is made twice. Three cases answer409with the codeIDEMPOTENCY_CONFLICTandrequest_repeatedas itsdetails.reason. While the first request is still running, send it again shortly to get its answer; a start, an upload, a send, or a create whose first request was cut off can answer this for up to 30 minutes. When the first request failed and Point11 can't tell whether it made its change, read the item to see where it stands, and use a new key if the change still needs making. When the first answer was larger than 65,536 bytes, it was not kept, so read the item to see where it stands. An answer that holds a link only you should use, such as an upload link or a live sign-in view, is never kept either: send that request again with a new key to get a fresh link. Use a new key for each new request. Every request that changes something needs a key, and one sent without it is refused. Through the MCP server, your assistant sends the key asidempotencyKey, which every tool that changes something requires. - Lists. Pass
limit, up to 100. To get the next page, pass thenextCursorfrom the previous response ascursor.nextCursorisnullon the last page. - Longer work. Starting a scan or a campaign answers
202right away with arunId. Check its progress withGET /api/v1/platform/runs-v2/{runId}, orplatform_runs_get_v2in the MCP server. - Errors. Every API error has a
code, amessage, and arequest_id. Decide what to do from thecode, and include therequest_idwhen you ask for help. Through the MCP server, a tool that refuses a request answers with a result marked as an error, not an HTTP status, and its text is the samecode,message, andrequest_id. When you don't approve a change, the code isCONFIRMATION_REQUIREDand nothing changes. - Limits. Each key has a request budget. When you sign in, each assistant or app you connect has its own budget, shared across your organizations, so one assistant never uses up another's. In the API, the
RateLimit-RemainingandRateLimit-Resetheaders show what's left, and when it runs out you get429; wait theRetry-Afterseconds, then try again. Through the MCP server, a tool answers a used-up budget as an error result with the codeRATE_LIMITEDandretryAfterSeconds, the seconds to wait.
Use the CLI#
The CLI is Restish, which reads the API reference and turns each request into a command named after it, such as discovery-campaigns-list. It uses an API key. Install Restish, for example with brew install restish, put your key in POINT11_API_KEY, then connect Restish to Point11:
restish api connect p11 https://point11.ai --no-discover
restish api set p11 'profiles.default.auth: {type: bearer, params: {token: env:POINT11_API_KEY}}'
restish api set p11 'profiles.default.credentials.apiKey.auth: {type: bearer, params: {token: env:POINT11_API_KEY}}'
restish api sync p11Then run any request by its command name:
restish p11 discovery-campaigns-listFor a request that changes something, put the Idempotency-Key first:
restish p11 inspector-scans-start "$(uuidgen)" url: https://example.comRun restish api sync p11 again to pick up new requests.