The server determines who charges
Check the endpoint before applying a price table. These connections serve different requests.
The official X setup guide documents app-only Bearer access for read tools and an OAuth bridge for user-context operations. CarryFeed's setup uses CarryFeed OAuth and your website account.
| Connection | Use it for | Cost to check |
|---|---|---|
| https://api.x.com/mcp | Official X API operations. | X developer-app credits and the operations called. |
| https://docs.x.com/mcp | Searching and reading X developer documentation. | Your AI client's costs; these tools retrieve documentation. |
| https://api.carryfeed.com/mcp | Supported public social sources, X search, source metadata, and media tools. | Free public X lookups; Tools balance or prepaid Instagram lookups for metered tools; separate AI client costs. |
CarryFeed tool costs
CarryFeed’s public X lookup and search tools are free. Reading post text or listing available X media metadata is different from creating a metered media result.
Connect through CarryFeed OAuth, then call `get_tool_usage` with `{}` to read the account’s Tools balance, prepaid Instagram lookups, and current prices. The media pricing page lists the supported results and limits. No monthly subscription is required for these media tools.
- Read the current tool list and price before requesting a metered result.
- Set `maxCostMicros` to the approved per-request limit; this is not a budget for the whole research session.
- Give each metered task a unique `requestId`. Reuse it with identical input on a network or billing_pending retry.
- Failed requests release their reservation. A new request ID represents a new task, not an automatic free retry.
| Task | Current tools | Billing boundary |
|---|---|---|
| Read or find a public X source | get_public_x_post, resolve_public_x_input, search_public_x_posts, list_public_x_media | Free public source lookup or metadata; the AI client may charge for its own work. |
| Read a public Instagram profile, photo post or Reel | resolve_public_instagram_input | A successful result uses a prepaid Instagram lookup first, then the Tools balance. A new request can be charged even when served from cache. |
| Extract YouTube captions or prepare a public video file | get_youtube_transcript, download_youtube_video, download_tiktok_video | Metered results use the Tools balance. Check the returned unit price and the tool’s source and file limits. |
Official X read prices
X's published pricing charges reads by returned resource. These USD rates apply to X API usage, not to every provider offering an MCP connection.
Owned Read pricing has endpoint and ownership conditions. Use the rates shown for your account in the Developer Console when estimating a run.
These are X’s published USD rates checked on October 31, 2026. Confirm the rate for your app in the Developer Console before running the estimate; they do not set CarryFeed’s media prices.
| Read operation | Unit price | Example |
|---|---|---|
| Post read | $0.005 per resource | 100 newly billed posts = $0.50 |
| User read | $0.010 per resource | 25 newly billed profiles = $0.25 |
| Qualifying Owned Read | $0.001 per resource | 1,000 qualifying resources = $1.00 |
A $0.75 research run
Suppose an agent collects discussion of a product launch. It retrieves 100 distinct posts and 25 distinct author profiles, all newly billable at the standard rates. The estimate is `(100 × $0.005) + (25 × $0.010) = $0.75`.
If you repeat that workload on 30 separate UTC days, with every resource billable each day, the estimate is $22.50. Add the AI client's costs.
Count returned resources across all calls. Pagination, author lookups, and following a thread can increase the total. Reuse saved text when the next task only needs another summary of the same sources.
- Choose one topic and a date range.
- Set a maximum of 100 distinct posts and 25 profile lookups for this example.
- Stop retrieval at those limits and draft from the saved sources.
- Compare the estimate with the actual usage before increasing the workload.
Set the budget before the agent starts
X's daily deduplication rules normally charge a resource once within a UTC day; the window resets at midnight UTC. Deduplication is a soft guarantee, so a local cache remains useful.
Set a spending limit in the Developer Console. The billing documentation explains the usage views and a monthly pay-per-use cap of 3 million Post reads. Higher volume requires Enterprise access.
An empty response and a large successful response have different usage. Failed requests that return no data are not billed; successful reads containing resources contribute to consumption.
Handle request limits separately
An API request can return many posts. Billing counts the returned resources, while endpoint rate limits count requests in a time window.
A 429 response can arrive while credits remain. Use `x-rate-limit-remaining` and `x-rate-limit-reset` for the endpoint your tool calls. Queue follow-up requests until the window resets.
For CarryFeed’s current media charges, see media tool prices. For source fields returned to an agent, see the metadata guide.
- Use the Per App limit for app-only access and the Per User limit for an OAuth user.
- Reuse saved results instead of re-fetching a source for each draft.
- Keep the task's data budget separate from the client's model budget.