Does X have an official Ads MCP server?

Yes — X runs one at ads-api.x.com/mcp over Streamable HTTP. It listed 74 tools for our token on 2 October 2026, and on X's Ads Starter plan it is the route that works: the REST Ads API answered the same token with an empty 403.

How do I connect Claude to X Ads?

Either point your MCP client at X's own server with your own enrolled developer app and OAuth token, or use a hosted server: connect X Ads under Integrations, add the endpoint to Claude Code, Claude Desktop or Codex with an API key, and scope each call to an ads account from x_ads__list_accounts.

Can an MCP client delete X Ads campaigns?

No. X's Ads MCP has no campaign or line-item delete, so pausing is the stop. Through this server, the deletes X does offer — audiences, cards, tags, media — and removing a targeting criterion are held for a person on every key mode.

MCP

X Ads MCP server: X's own server, and what a hosted layer on top of it adds

TL;DR

X Ads has an official MCP server (ads-api.x.com/mcp), and on X's Ads Starter plan it is the only route that works: our first live connect got an empty 403 from the REST Ads API and 74 tools from the MCP server. This guide covers using X's server directly, and what a hosted layer adds for unattended or multi-platform work.

What already exists for this platform

X publishes its own Ads MCP server over Streamable HTTP. It is stateless — a tool call needs no session — and answers in server-sent events. With an X developer app enrolled in the Ads project and an OAuth 2.0 token carrying ads.read and ads.write, any MCP client can call its tools directly.

An official server, 74 tools at the time of our connect: accounts, funding, campaigns, line items, promoted posts, targeting, stats, audiences, cards, pixels, media.
Your own developer app. X's developer console puts the Ads product on the "Ads Starter" plan, described as MCP access only — the REST Ads API answered our token with an empty 403.
Raw units and raw limits. Amounts are micro-units, stats are seven days and twenty ids per call, and nothing stops a call from targeting an account you did not mean.

What a hosted layer adds

	X's Ads MCP directly	Agent Planners MCP
X Ads tools	X's tool list as-is	10 reads, 8 writes, plus any other server tool by name
Account binding	Your own discipline	Every call names the selected ads account
Money	Micro-units	Whole account currency in and out, converted once
Stats beyond 7 days / 20 ids	Split it yourself	Split into windows and merged, in the account's time zone
Creates	Whatever you send	Always created paused
Permanent deletes, removing targeting	Applied when called	Held for a person on every key mode
Other platforms in the same session	No	28 more platforms on one endpoint

Counted from the tool registry; tools/list offers the same 18 X Ads tools of the 1012 on the endpoint (a key that lists tools on demand reaches them through search_tools and call_tool). A key only sees the writes if it carries tools:write.

How to configure it

1Create a key in API & MCP. tools:read for reporting; add tools:write for changes. Tools mode needs a paid plan and fails closed.
2Connect X Ads under Integrations, signing in with an X account that has a role on your ads account (ads.x.com → Settings → Users). The MCP key never carries X credentials.
3Add the server to Claude Code, Claude Desktop or Codex with the snippet below.
4Run describe_permissions, then x_ads__list_accounts for the ads account ids, and scope every call with one.

bash

claude mcp add --transport http agentplanners https://www.agentplanners.com/api/mcp \
  --header "Authorization: Bearer ap_live_…"

Codex uses the same endpoint from ~/.codex/config.toml, with the key in an environment variable rather than the file: [mcp_servers.agentplanners] / url = "…/api/mcp" / bearer_token_env_var = "AGENTPLANNERS_API_KEY".

What waits for a person, whatever the key allows

A tools:write key's writes either run inside the organization's guardrails or queue for an admin, depending on how the key was minted. One class never runs on the key's say-so: it is held in the approval queue, an accountable admin or editor is emailed, and it is decided on its own.

The server's permanent deletes — a custom audience, a card, a web or app event tag, a tracking tag, a media item or media creative, a do-not-reach list, an app list.
Removing a targeting criterion. A line item left with no criteria serves broadly, so taking one away can widen spend.

A status of DELETED and any delete-shaped write outside that list are refused outright. There is no campaign or line-item delete on X's server — pause is the stop.

X's ad model, for anyone building on it

Campaigns carry no budget. create_campaign takes a name and a funding instrument; daily and total budget, bid, schedule, pacing and placements belong to the line item, and a line item needs a daily budget and a start time.
Pause and activate are separate. An update can pause a campaign but not activate it; activation is its own call, and a line item serves only when its campaign is active as well.
Placements for stats are ALL_ON_TWITTER or PUBLISHER_NETWORK, and stats days follow the ads account's time zone.

Frequently asked questions

Does X have an official Ads MCP server?: Yes — X runs one at ads-api.x.com/mcp over Streamable HTTP. It listed 74 tools for our token on 2 October 2026, and on X's Ads Starter plan it is the route that works: the REST Ads API answered the same token with an empty 403.
How do I connect Claude to X Ads?: Either point your MCP client at X's own server with your own enrolled developer app and OAuth token, or use a hosted server: connect X Ads under Integrations, add the endpoint to Claude Code, Claude Desktop or Codex with an API key, and scope each call to an ads account from x_ads__list_accounts.
Can an MCP client delete X Ads campaigns?: No. X's Ads MCP has no campaign or line-item delete, so pausing is the stop. Through this server, the deletes X does offer — audiences, cards, tags, media — and removing a targeting criterion are held for a person on every key mode.