SDK plan limits

The MCP Studio SDK has a Free plan and an SDK Pro plan.

Audience	SDK buyers, platform admins, finance stakeholders
Prerequisites	Access to SDK pricing details and expected monthly MCP/server usage

Current limits

Your plan sets two limits, and they are counted differently. Read this before the table, because the difference is the thing people get wrong:

• Servers are counted per end user. Each person who creates a server through your embed gets their own allowance. One customer reaching it does not affect anyone else, and there is no cap on how many end users you have.
• Requests are counted across your whole integration. Every request to every server created through your embed, by every end user, draws down one shared monthly allowance. This is the limit that runs out.

Resource	Limit	Counted
Free plan MCP servers	1	per end user
SDK Pro MCP servers	10	per end user
Free plan requests	1,000 MCP requests/month	across all end users
SDK Pro requests	5,000 MCP requests/month	across all end users
Sources per server	2	per server
Indexed chunks per server	Up to 500K on Free, up to 3M on SDK Pro	per server
SDK integrations per developer	No fixed cap	—
Widget configurations	No fixed cap	—
End users	No fixed cap	—

Your end users also have their own Context MCP Studio account limits, which apply on top of yours: 3 servers and 50 requests per month on the standalone free tier, with the 50 requests shared across all of that person's servers rather than counted per server. So a request can be refused either because your integration spent its monthly allowance or because that individual spent theirs, and the message tells them which.

MCP calls

An MCP call is a single request from an AI agent (Cursor, Claude Desktop, Windsurf, etc.) to an MCP server endpoint. Each tool invocation — for example, searching docs, extracting code, or asking a question — counts as one call.

Source limits

Each MCP server created by your end users can connect up to 2 included sources. Sources include websites, GitHub repos, documentation sites, API endpoints, PDFs, or other MCP servers. The same source guidelines from MCP Studio apply:

• Large websites can take longer indexing windows depending on crawl size and quality. Indexing continues in the background until the source completes or fails.
• Private GitHub repos require authentication from the end user.
• Source content is periodically refreshed. Unchanged pages skip re-embedding.

Source size

Beyond the source count, each MCP server has a cumulative allowance of indexed chunks shared across all of its sources. A chunk is a small passage of content stored so an AI agent can retrieve the exact part that answers a question.

Plan	Indexed chunks per server
SDK Free	Up to 500K
SDK Pro	Up to 3M

When a server reaches its allowance, indexing stops at the limit rather than failing. Everything already indexed stays queryable, and the end user is told that a larger allowance is available on a higher plan.

Limits reached

• Server limit: The end user is told your app allows this many servers per user, that they can delete one to create another, and that it is not a limit on their own account. Your other end users are unaffected.
• Source limit: After 2 included sources, each additional source is billed at $3.
• Source size limit: Indexing stops at the plan allowance; existing content stays queryable and upgrading raises the allowance.
• Request limit: Requests are refused for the rest of the calendar month once your shared allowance is spent, and the end user is told it is your app's allowance rather than theirs. Because this limit is shared, it affects every end user at once — it is the one to watch.

Both the server and request limits are hard stops unless you turn on usage beyond your plan, below.

Usage beyond your plan

By default the limits above stop your end users. If you would rather keep serving them and pay for it, turn on usage beyond your plan in the Developer Portal and set a monthly budget:

• $0.10 per additional MCP request
• $25/month for each MCP server beyond a single end user's allowance

Two things are worth understanding before you enable it.

The budget is required, and it is the most this can add to your bill. Requests and extra servers draw on the same figure. Once the month's overage reaches it, we stop serving overage and your end users are refused exactly as they were before — in your app's name, with no mention of your budget or your bill.

Extra servers are priced once a day, not the moment one is created. A server created and deleted the same day is not billed, and the charge falls again when your end users delete servers.

Turning it off cancels the billing straight away. Usage already served that month is still invoiced. If a payment for it fails, we follow the same 14-day recovery window as any other subscription and then switch overage off, which puts your limits back to hard stops rather than leaving them unpaid.

Analytics dashboard subscriptions

New SDK developers get a 30-day trial with Core, Action, and Predictive included. After the trial:

• Core Analytics: $350/mo per MCP server
• Core + Action: $450/mo per MCP server
• Core + Action + Predictive: $550/mo per MCP server

You subscribe once at the integration level, and the tier is granted to every end user who deploys a server through your embed — they do not buy an MCP Studio analytics subscription separately.

Each tier also unlocks matching depth in your own business intelligence dashboard at Portal → Embed Usage:

Tier	Portal insight added
Core	Adoption and activation rates, dormant servers, AI client mix, source reach
Action	Top questions across all end users, aggregate content gaps, ranked fixes
Predictive	Requests-to-answer efficiency, aggregate source grades, forecasts, at-risk integrations

This is why SDK analytics is priced above MCP Studio standalone: it reports at the level of a business embedding the product, not a single server. See Analytics and observability.

Questions

For questions about limits, pricing, or enterprise needs, contact zair@appatools.com.