How-To Guide
How do you add Blotato to Google Antigravity?
Add a social media MCP server to Google Antigravity: the config file Google documents, the verification read, and what a connected agent can and cannot do.
By the Superdirector editorial team. Google's Antigravity MCP documentation, the MCP documentation site, Blotato's blog, client config and help pages, and Superdirector's MCP and pricing pages were read November 15, 2026.
At a glance
How do you add Blotato to Google Antigravity?
Blotato's own answer, quoted here because it is theirs: paste one JSON config block with your API key in the header into Antigravity's mcp_config.json, then restart the app. Google's side of that is documented too: mcp_config.json lives at one global path for the IDE and the CLI, remote servers must use the Streamable HTTP transport, and unconfigured MCP tools run in Ask mode that wants your approval before a tool executes. The same protocol carries any social MCP: Superdirector's MCP (ours) connects with one endpoint and a browser sign-in, reads your saved brand context, and leaves drafts in your Canvas; it does not post from the chat.
- Antigravity supports MCP through one config file read by the IDE, the App and the CLI; the endpoint goes in serverUrl and the connection must be Streamable HTTP.
- A key sits in a headers block; OAuth servers sign in through the browser instead. Antigravity asks for approval before unconfigured MCP tools run.
- MCP connects the accounts; it does not decide publishing. Blotato's server publishes from their queue per their docs, while Superdirector's drafts and posts nothing from the chat.
Three menus, one file: whichever surface you use, the same config carries the endpoint and the transport. Original diagram, drawn from Google's Antigravity MCP documentation and Blotato's walkthrough, both read November 15, 2026.
Make it work for your brand
Find it. Remake it. Post it. Superdirector turns the viral formats in your niche into your brand's carousels, X posts and video scripts, and keeps the calendar full.
What an MCP setup means before you paste anything
Google's documentation opens with the sentence the whole setup rests on: "Antigravity supports the Model Context Protocol (MCP), an open standard that lets AI agents and editors securely connect to local developer tools, databases, file parsers, and external remote APIs." The standard's own documentation puts it more plainly still: "Think of MCP like a USB-C port for AI applications." One connector shape, many servers, one place where credentials live.
What the setup adds to your agent is worth saying precisely. Antigravity has no platform credentials of its own, and even the vendor that sells the connection says so: Blotato's FAQ states that "Google Antigravity cannot reach a social account on its own, because it has no platform credentials." The credential lives in one place, on the server you connect, and everything the agent drafts travels through that server's tools.
Two boundaries follow before any config exists. First, permission is on by default: Google documents that "unconfigured MCP tools run in Ask mode, requiring your approval before execution," which means no social tool runs unattended the moment you connect it. Second, a connection is not a publishing decision: whichever server you connect, the server holds the accounts and the agent holds the words, and the two only meet when a tool call is approved.
Get the endpoint and the credential from the server you chose
Every remote MCP setup starts inside the server's own app, not inside Antigravity, and the two routes below differ exactly where the credential lives.
Blotato's route is a key. Their walkthrough says to sign in and "open Settings > API . Click Generate API Key if you don't have one yet, then under "Other MCP Clients" click Copy MCP Config," which copies a ready-made block with the key already inside, so a header is never assembled by hand. Their help center states the plan condition in their own words: "Confirm your paid Blotato subscription. API and MCP access is not included in the free trial," and their config page warns that "generating a key activates the paid Starter plan." Those are their terms on their pages, and they matter before you paste anything, because the trial excludes exactly the access the MCP needs.
Superdirector's route (ours) is a browser sign-in instead of a key. The manual setup page shows the endpoint, "Remote MCP endpoint https://superdirector.app/mcp," and states that you can "Connect and check your account for free." In Claude, the guide hands you one link that opens "Claude's own connector dialog prefilled with this environment's /mcp endpoint (URL-encoded)"; in Claude Code it is the command "claude mcp add --transport http superdirector "https://superdirector.app/mcp"" plus a browser login; in Codex it is an entry in ~/.codex/config.toml with its matching login command. The consent screen that follows names what it covers: "all current and future Canvases" in your account, so read it before you click approve.
Which role each server plays matters more than the mechanics. Blotato's own framing is "Antigravity does the writing, Blotato does the publishing." Superdirector's MCP connection is deliberately narrower: per its guide, "the AI reads the Canvas's sources and brand context, saves drafts, and starts credit-using work only after a credit quote and the person's OK," and "Posting is approved in the Canvas, not from the chat." Both connect through the same protocol; what differs is what the server does.
Paste the config into Antigravity
Antigravity reads one config file across its surfaces, and where to open it depends on which one you run. In the App, Google documents the path as "Click Customizations in the sidebar. Click Installed and scroll to the MCP Servers section." In the IDE, it is the agent side panel's menu: "Click Manage MCP Servers . Click View raw config ." In the CLI, "Type /mcp inside the prompt panel and press Enter to open the interactive MCP Manager Overlay ." All three read the same file: the global path is ~/.gemini/config/mcp_config.json, and a workspace copy at .agents/mcp_config.json applies only to that workspace per Google's doc.
The block both servers fit looks like this, with each server's own endpoint and each credential exactly as that server documents it:
{
"mcpServers": {
"blotato": {
"serverUrl": "https://mcp.blotato.com/mcp",
"headers": {
"blotato-api-key": "YOUR_BLOTATO_API_KEY"
}
},
"superdirector": {
"serverUrl": "https://superdirector.app/mcp"
}
}
}Four field rules from Google's doc decide whether this connects. Remote servers "must use the Streamable HTTP transport," and the legacy HTTP+SSE transport and the httpUrl field are not supported. The endpoint belongs in serverUrl, and Google's reference notes that "url also works" as the field name, though Blotato's walkthrough reports a build where a bare url entry failed with "serverUrl or command must be specified," so serverUrl is the spelling every stored page agrees on. Headers are the documented home for key authentication: "Custom HTTP request headers sent with remote serverUrl connections (for example, Authorization or API key headers)." And for servers that support modern OAuth discovery, Antigravity "automatically handles OAuth 2.0 with Proof Key for Code Exchange (PKCE)," which is the path Superdirector's endpoint uses, no header needed.
One more habit worth copying from the vendor whose server it is: Blotato's page says to expect the httpUrl field to be the first thing that breaks a setup copied from elsewhere, because "The first is the httpUrl field, which Gemini CLI uses and Antigravity does not support, so a block copied from Gemini CLI does not register." If a config arrives from another client's docs, diff the fields before pasting, not after an error.
Load the server and verify with one read
Blotato's doc describes the reload behavior: "Restart Antigravity after every config edit, and validate the JSON at jsonlint.com if it still fails." Their config page adds the reason in their own words: servers load on startup, so a refresh button can lag behind a working connection.
Then make one cheap read instead of trusting a badge. Blotato's pages converge on a single verification question, "ask your client to list connected Blotato accounts," and define expected success tightly: "An empty array is a valid result when you have no connected accounts." That detail is the honest check most guides skip: an empty list proves the connection and authentication work, and it also proves no accounts are connected yet, which is a setup step on the server's side, not a bug. Their help center is blunt about the counterfeit: "A Connected badge alone does not verify a tool call."
Superdirector's equivalent read is documented in its guide: "Call the intended server's native account_status and confirm connection.access === 'all_canvases'", then list canvases and credit balance from the same native reads. The consent screen you approved is the record of what the connection may reach, so the verification read should match what that screen named.
What the connected agent can and cannot do
Blotato's stated scope, on their pages: connecting to their hosted MCP server "gives it tools to publish and schedule across nine platforms: X, LinkedIn, Facebook, Instagram, TikTok, YouTube, Threads, Bluesky, and Pinterest," their walkthrough frames the split as "Antigravity does the writing, Blotato does the publishing," and their FAQ ties a scheduled post to their own queue: "Scheduling runs on Blotato rather than inside Google Antigravity." They also carry their own metrics caveat on their config page, that analytics through MCP cover some platforms while TikTok, YouTube and Pinterest metrics live in their web app and API and LinkedIn reports none yet, which is their stated limit, worth reading before you promise anyone a report.
Whatever the server promises, Antigravity's own permission default still applies to it: "unconfigured MCP tools run in Ask mode, requiring your approval before execution." A posting tool connected today runs with your approval the first time, which is where the guide-shaped promises of unattended posting meet Google's actual default.
- The agent side drafts, grades, and prepares prompts against what is connected.
- The MCP server holds the connection: endpoints, credentials, connected accounts, and whatever tools the vendor documents.
- The account holder decides what publishes, on whichever side owns that decision: their queue, your Canvas, your approval.
Superdirector's MCP connection (ours) sits entirely on the drafting side of that table, and says so itself: "In connected AI apps, Superdirector plans and prepares; images and storyboards are made in the Superdirector web app when the person presses Create, and this connection returns no images or media files." It reads your saved brand context and sources, saves drafts to your Canvas, and runs paid work only after a quoted cost and your approval. If you wanted an agent that posts by itself, this connection is not it, and the section below names what is.
The agent drafts, the fence asks first, the server holds the connection, and nothing crosses to a platform on its own. Original diagram, drawn from the article's reading of Google's Antigravity MCP documentation and Blotato's pages on November 15, 2026.
Common issues and fixes
Both sides' docs name the same small set of failures, so the table keys each one by the sign you actually see, not by what a tutorial says should happen:
| The sign you see | The documented cause | The check that settles it |
|---|---|---|
| 401 "invalid API key" | Blotato's help page: a copied key missing its trailing = characters causes exactly this error | Re-copy the whole key from their Settings, including trailing equals signs, and keep it out of shared project config |
| Server never appears in the list | A JSON error, a wrong field, or an un-reloaded config; Blotato's pages report servers load on startup | Restart after every edit, then confirm the block uses serverUrl and Streamable HTTP, not httpUrl or legacy HTTP+SSE |
| A block copied from Gemini CLI does nothing | Blotato's config page: the httpUrl field is not supported in Antigravity | Rename the field to serverUrl and check the transport before re-adding the server |
| A tool call returns 403 | Blotato's troubleshooting separates the two: "A subscription-related 403 requires an active paid plan. A 401 points to credentials or session authentication" | Plan-side: confirm the server's paid tier is active; credential-side: re-authenticate in the browser |
| Tools connect but nothing useful runs | The accounts side is empty: an empty account list from the verification call is a valid success | Connect a social account in the server's own app first, then re-ask the listing question |
| Paid actions refuse after connecting | Every stored path asks for consent somewhere: Blotato in their browser authorization, Superdirector with a quoted cost and your OK per run | Read the actual refusal message; it names whether the missing piece is a plan, a permission, or a credit decision |
If you would rather the account runs without you in the seat
This page sells Superdirector, so its offers are stated the way the pricing page states them, and the two $29 plans already met above stay distinct. The $29 Starter is the MCP access plan: "If you work from Claude or ChatGPT, Starter at $29/month is all you need: live social data, ideas, scripts and downloads." The $199 Growth is where scheduled operation lives: "Choose Growth at $199/month for scheduled research, routines and supported Instagram and X publishing." Growth is one AI-run operated account on the $199 plan; the pricing page rows include two accounts on it and put extra accounts at $99 each per month.
The publishing boundary on Growth is account-level, not per post: it "publishes eligible Instagram carousels and single X text posts by themselves once posting is turned on for an account," and that switch comes "after one warning that names the account, the public effect and the cost." The MCP connection above never posts from a chat regardless of plan: drafts are approved in the Canvas.
Two honesty lines from the same page close this section. Credits "cover AI work and vary by task; they are not a count of finished videos," so a plan with more credits is not a promise of more posts. And the decisions stay with you: "You make the important calls: what posts, what changes and what gets paid."
Where this setup ends
A queued post is not a published one. Blotato's own FAQ distinguishes an agent that drafts from a system that runs unattended, and even their publishing queue publishes while nothing on this page can show platform readback, so no sentence here claims any post went live. Verification of a publication lives in the platform, not in the agent pane.
Labeling follows each platform's rule, and the pricing page states the house version: "Label AI content where a platform requires it, and never publish invented reviews, testimonials, results or customers." That rule binds this page too: every view count, review score and self-reported result that appeared in the source pages read for this article belongs to the vendor who reported it, and none of it is repeated here as a result you should expect.
Vendor terms also move: Blotato's trial and plan conditions are stated on their pages as of November 15, 2026, Google's config schema is documented on theirs, and a reader's installed build may sit before or after either. The durable part of this setup is the shape: one config file, serverUrl and Streamable HTTP, a credential where the server documents it, one verification read, and a person somewhere in the approval loop.