X Hosted MCP at $0.01/Call: A New Baseline for Agent Data Access Costs

What X Launched

On June 30, 2026, X (Twitter) launched its hosted MCP server, letting AI agents talk to the X API via the Model Context Protocol without setting up their own OAuth flow or proxy. Setup steps: create an X developer app, top up billing, obtain a config ID, and connect it in Claude Code, Cursor, Codex, or any MCP-compatible agent host.

The pricing hook: personal tier $0.01 per API call, or 1,000 calls for $1. Early users reported pulling three days of bookmarks for $0.10. That is the number to anchor on when comparing MCP costs across platforms.

MCP Pricing Across Platforms

MCP isn't a pricing model — it's a protocol. What matters is what the platform behind the MCP server charges for the underlying API access:

X's per-call pricing looks expensive on this table, but the comparison is misleading. GitHub's free tier caps at ~5,000 requests/hour and Slack caps by tier. X's $0.01 is uncapped — the meter just keeps running. For low-volume research agents this is fine. For a monitoring agent polling every 10 seconds, it's a bill.

Agent Workload Cost Scenarios

Scenario 1: Research agent building a weekly digest

A research agent pulls tweets from 20 accounts, weekly. That's ~500 API calls per run, ~$5/week or $20/month. Cheap. This is the intended use case.

Scenario 2: Sentiment monitoring on a brand

Monitor mentions of “Cursor” every 5 minutes with search + pull. That's ~576 calls/day, ~$5.76/day, or $173/month. Real money — comparable to a Cursor Team seat itself.

Scenario 3: Full-firehose social listening

For real-time firehose access, the per-call model breaks — you'd need X's Enterprise API tier, which starts at $5K/month and doesn't bill per call. Hosted MCP is not the right entry point for this use case.

Combined Cost with the LLM

An agent that calls X and processes tweets in the same session pays two meters: the MCP call fee and the LLM token cost. Rough estimate for the research digest scenario:

• X MCP calls: $5/week
• Claude Sonnet 5 promo (analyzing tweet content): ~$8/week at 2M input, 100K output tokens/session
• Total: ~$52/month for a weekly digest workflow

The MCP share is 24% of total agent cost. Meaningful but not dominant. If X had priced this at 5× the current level, MCP would become the cost driver.

What This Signals for MCP Pricing

X's $0.01/call is a psychological anchor. Historically MCP servers have been either free (extension of existing free APIs) or bundled into paid platforms. A per-call price on top of an API access fee is new. Other platforms watching:

• Reddit — has expensive API access but no hosted MCP; per-call pricing could restore agent access.
• LinkedIn — could unbundle Sales Navigator into MCP metering for agents specifically.
• Stack Overflow — currently free but rate-limited; a paid MCP tier would be viable.

The likely direction: platforms that lost publisher-side value to LLMs get another revenue stream from agent-side data access. Expect $0.005–0.02 per call to become the norm for social-data MCPs by end of 2026.

Practical Budgeting Guidance

If you're building an agent that touches X data:

1. Cap the agent's per-run call budget explicitly in the MCP config.
2. Use batched search queries instead of per-user pulls where possible.
3. Cache results locally — MCP calls to X for the same query within an hour are almost always redundant.
4. Track MCP spend and LLM token spend as separate line items so you know which is scaling.

The $0.01 per call sounds trivial; that's exactly when unbudgeted agents run 100K calls overnight. Treat MCP calls the same way you treat LLM tokens: measured, capped, alerted.