Antigravity vs Cursor: Plans, Models and a Twitter (X) MCP Setup in Both

The short version: Antigravity is Google's agent-first platform and leans on Gemini, with a real free tier; Cursor is now a SpaceX product bundled with Grok and X, with its own models in the subscription and everyone else's models billed by usage. Both accept MCP servers through almost the same JSON file.

How to read the table. Antigravity's own pricing page lists the plans (Individual $0, Google AI Pro, Google AI Ultra, Business via Google Cloud) without printing the Pro and Ultra prices; Google sells those through one.google.com/about/google-ai-plans, where the price depends on country, so check there for your region. The Cursor prices come from cursor.com/docs/models-and-pricing, and the Claude Code prices from claude.com/pricing, where Claude Code is included in Pro ($20 a month, or $17 a month billed annually), Max ($100) and Ultra ($200) but not in the free plan.

Antigravity. The plans page says every plan gets Gemini 3.8 Flash as a core agent model, unlimited Tab completions and access to all product features, including scheduled tasks and the CLI. Accounts without Google AI Pro or Ultra get "a meaningful quota, refreshed weekly" plus a weekly rate limit. Pro gets a quota refreshed every five hours until a weekly limit is reached; Ultra gets the highest quota, also refreshed every five hours, and access to third-party models. Google states that the limits track how much work the agent does, so short tasks stretch further than long ones, and that the limits can change.

Antigravity overages. Pro and Ultra users can spend purchased Google AI credits once the baseline quota runs out, at Gemini Enterprise consumption pricing, controlled by an AI Credit Overages setting (Never or Always). The same page says there is no bring-your-own-key option and no custom-contract organizational tier for individuals; teams go through Gemini Enterprise, where the pricing page lists Standard and Plus plans from $30 per seat per month and a pay-as-you-go plan with a $0 seat fee.

Cursor. Since the SpaceX acquisition, Cursor is sold as part of The X Subscription, one subscription for Grok, Cursor and X. Free has limited Agent and Chat usage. Plus at $8 a month runs only Cursor's own models (Grok 4.7, Grok 4.6, Grok 4.5 and Composer 2.5) at fixed settings and excludes third-party models and your own API keys. Premium ($30), Super ($100) and Ultra ($200) give 4x, 15x and 50x the Plus usage, and Teams seats cost $30, $100 or $200 per seat a month. Annual billing is $80, $300, $1,000 and $2,000 a year for the four paid individual tiers.

The pool is shared and weekly. Cursor's docs say each paid plan includes one pool for Cursor models, shared across Agent in the editor, the CLI, Cloud Agents, Grok Bot and the Grok apps, refreshed weekly. Claude, GPT and Gemini are outside that pool: on Premium and above they bill at the model's API list price from the first request, paid from prepaid credit with an optional monthly cap. If your team standardizes on Claude or Gemini models, budget for that usage separately from the subscription.

Antigravity's models page lists which reasoning models each plan can pick, with a thinking level of Low, Medium or High. Gemini 3.8 Flash is available on every plan. Gemini 4 Argon needs paid Google AI Pro or Ultra. Claude Sonnet 5.5 and Claude Opus 5.5 (thinking) are available on Ultra and on non-trial Pro subscriptions. The page also marks Claude Sonnet 4.6, Claude Opus 4.6 and GPT-OSS-120b for removal on November 2, 2026, so the free tier's non-Gemini options were shrinking at the time of writing. Image generation inside Antigravity uses Nano Banana 2 and is not configurable.

Cursor's included models are its own: Grok 4.7, 4.6 and 4.5 plus Composer 2.5. Third-party models listed in its pricing docs include Claude Opus 5.5, Sonnet 5.5 and Haiku 5.5, GPT-5.6 variants and Gemini 3.1 Pro, 3.8 Flash and 4 Argon, all charged per token from prepaid credit on Premium and above. Teams and Enterprise pay an extra Cursor Token Rate of $0.25 per million tokens on eligible third-party models; individual plans do not.

Practical read. If you want Gemini 4 Argon or Claude 5.5 inside the subscription quota, Antigravity on Pro or Ultra covers it. If you are happy on Grok and Composer, Cursor Plus at $8 is the lowest paid Cursor tier. If you want Claude models without per-token billing in an editor, neither is the obvious fit and Claude Code on a Claude plan is the more direct route, which is why the third column keeps showing up.

Both tools read a JSON object with an mcpServers map, so a server definition moves between them with one edit. Antigravity's docs name serverUrl as the field for a remote Streamable HTTP server (and accept url as an alias), and Cursor's docs use url. Both send custom headers with remote requests, which is all a Bearer-key server needs.

Antigravity specifics. In the IDE, open the ... menu at the top of the agent side panel, choose MCP Servers, then Manage MCP Servers and View raw config. The CLI has an interactive /mcp manager. Each server can carry enabledTools or disabledTools, and timeoutSeconds (default 300 seconds to connect and 180 seconds per tool call). Unconfigured MCP tools run in Ask mode, and you can allow them with patterns such as mcp(twitterapi-io/*). OAuth tokens are stored in ~/.gemini/antigravity/mcp_oauth_tokens.json.

Cursor specifics. Cursor resolves ${env:NAME}, ${userHome} and ${workspaceFolder} inside command, args, env, url and headers, so you can keep the key out of the file. It supports MCP tools, prompts, resources, roots, elicitation and the MCP Apps extension, and static OAuth client credentials for servers without dynamic client registration. Team admins can distribute MCP servers through a team marketplace.

TwitterAPI.io hosts an MCP server at https://mcp.twitterapi.io/mcp. Its initialize response, which we requested this session, describes 12 tools (tweet search, user lookup, followers, replies, trends and more) and says to paste your TwitterAPI.io API key as the Bearer token. You get a key from a Google sign-in, with no X developer account; new accounts that pass the automatic bot check get up to $1 of free credits (at most 100,000 credits, no card, one per person or network, expiring after 31 days).

Step 1. Copy your API key from the dashboard. Step 2. In Antigravity, open View raw config (IDE) or edit ~/.gemini/config/mcp_config.json directly; use <workspace>/.agents/mcp_config.json if only one project should see the server. Step 3. Paste the block below and save. Step 4. Back in MCP Servers, refresh and check that twitterapi-io lists its tools. Step 5. Optionally allow the read tools with mcp(twitterapi-io/*) so the agent does not stop for approval on every search.

The timeoutSeconds line is optional: 120 seconds is plenty for a paginated search, and a value set here replaces both the connect and the per-call limit. Do not commit a file that contains the raw key; the workspace file is meant to be shared, so keep keyed servers in the global file.

In Cursor, create or open ~/.cursor/mcp.json (all projects) or .cursor/mcp.json (one project) and add the block below. The only differences from Antigravity are the field name url and the ${env:...} interpolation, which reads the key from your environment so the file can be shared. Then open Customize and confirm the server shows as connected with its tools listed.

For Claude Code, one command registers the same endpoint; Claude Code's MCP docs list claude mcp add --transport http <name> <url>, with project-scoped servers stored in .mcp.json and local or user-scoped servers in ~/.claude.json.

claude mcp add --transport http twitterapi-io https://mcp.twitterapi.io/mcp --header "Authorization: Bearer YOUR_TWITTERAPI_IO_KEY"

Once connected, the prompts are identical in all three tools. Examples: "Search the latest 100 X posts mentioning antigravity and cursor, and tell me which other coding tools they name", or "Pull the last 40 posts from @cursor_ai and list every pricing or plan change they announce". The MCP tools return tweets as JSON, and each tweet returned is billed at 15 credits, so a 100-tweet answer costs about $0.015.

We searched X for recent posts containing both "antigravity" and "cursor" (Latest, reposts excluded) and kept the 364 posts whose text names both tools. They came from 317 distinct authors over about 9 days, roughly 40 posts a day. The striking result is that the comparison is rarely two-way: 59.1% of the posts also mention Codex, 57.1% mention Claude Code, and 67.6% mention at least one of them.

Money and limits come up, MCP less so. 18.4% of the posts use price, plan, limit, quota, credit or usage terms, 10.4% mention MCP, 8.5% talk about switching, moving or canceling, and 8.2% say "free". The audience is international: 59.1% of the posts are in English, 19.0% in Chinese and 10.4% in Japanese. Engagement is modest (median 1.5 likes, median 168 views, 33.2% with zero likes), which is what you expect from working developers rather than launch posts.

The official accounts tell you what each company wants you to notice. We pulled the most recent posts, replies included, from @antigravity, @cursor_ai and, for the third column, @claudeai. Cursor's account mentions pricing, plans or limits in 12.2% of its posts against 2.6% for Antigravity. Antigravity posts more often (1.35 a day vs 0.93) but spends most of that on conversation: 65.0% of its posts are replies and 21.2% are quotes.

Cursor's original posts get a median of 4,114 likes against 1,244 for Antigravity, on roughly 2.4x the followers. Per view, though, Antigravity's audience engages more: a median of 11.47 likes per 1,000 views vs 8.7 for Cursor and 5.56 for Claude. Video carries Antigravity's launches (59.6% of its original posts include video, against 37.0% for Cursor). MCP itself is a minor topic for all three accounts (2.0%, 2.6% and 0.6% of posts).

Question. What do the official Antigravity and Cursor accounts post about, and what else do developers bring up when they compare the two? Accounts and query. /twitter/user/last_tweets for @antigravity (160 posts over 118 days), @cursor_ai (120 over 128 days) and @claudeai (160 over 99 days), plus /twitter/tweet/advanced_search with antigravity cursor -filter:retweets (Latest, 400 posts returned, 364 naming both tools). Window. The most recent posts at collection time, in 2026. Method. Case-insensitive keyword matching on post text; topic shares count a post once per topic.

What the sample cannot show. Keyword matching misses sarcasm and posts that refer to a tool without naming it, the search covers about nine days only, and X posting is not the same as usage or market share. Engagement on official accounts reflects follower counts and paid promotion as much as product quality. Privacy. Aggregates only; the three named accounts are organization accounts, and search results were reduced to de-identified counts with no handles, ids or text kept.

Cost. The study took 46 requests and returned 840 tweets and 3 profiles, which costs about $0.13 on TwitterAPI.io (tweets at $0.15 per 1,000 returned, profiles at $0.18 per 1,000). The same reads on the official X API pay-per-use rate card at docs.x.com/x-api/getting-started/pricing ($0.005 per post read, $0.010 per user read) come to about $4.23, roughly 33x more per 1,000 tweets. Write actions (post, reply, like, DM) are also available through an authenticated session: a /twitter/user_login_v2 login ($0.10, using a proxy you supply) and then 60 credits ($0.0006) per call such as /twitter/create_tweet_v2. Those calls act as that account, so X can rate-limit, lock or suspend it; keep a human pace, expect occasional re-login, and note every HTTP 200 call is billed.

Which one to pick. Choose Antigravity if you want a usable free tier, Gemini 4 Argon or Claude 5.5 inside a subscription quota, and Google Cloud billing for a team. Choose Cursor if you live in a VS Code-style editor, are happy with Grok and Composer in the pool, or already pay for The X Subscription. Choose Claude Code if Claude models in a terminal are the point. For live X data the decision does not matter: the same MCP block works in all three. Grab a key from the dashboard, read the endpoint reference at docs.twitterapi.io, and paste the server into whichever tool wins.