GoHighLevel MCP server: what it does, which plan it needs and what it costs
Amber buttons and amber links are affiliate links: if you sign up through one, that vendor pays us a commission at no extra cost to you. Plain buttons and links are not. How we earn.
I am an independent entity from HighLevel. I am not an agent or employee of HighLevel and have no authority to make any binding commitment on behalf of HighLevel.
The GoHighLevel MCP server is HighLevel's OAuth connection that lets AI assistants such as Claude, ChatGPT and Cursor find CRM records and run supported actions, according to HighLevel's setup guide saved on 16 November 2026. That guide counts more than 600 operations across 38 domains, reached through 4 tools, and it names no plan requirement and no fee for the server. HighLevel's API article includes basic API access from the $97 Starter plan and reserves advanced access, OAuth included, for the $497 Agency Pro plan, so confirm with HighLevel which plan your connection needs. Texts an assistant sends and phone numbers it buys are not exempted from usage rates in either document.
What is the GoHighLevel MCP server, and where does HighLevel document it?
MCP stands for Model Context Protocol, a standard that gives AI assistants one way to discover and use the tools of a connected application. The GoHighLevel MCP server, which HighLevel also calls the LeadConnector MCP Server, applies that standard to the CRM. A connected assistant can look up records and carry out supported actions from a typed instruction, and nobody has to know the underlying API.
HighLevel documents the server in two places. The help center article “How to Setup and Use the HighLevel MCP Server”, modified on 15 November 2026, holds the setup steps for each AI client. The developer documentation at marketplace.gohighlevel.com/docs repeats the endpoints and tool list on a page titled “LeadConnector MCP Server”. This guide rests on both, saved on 16 November 2026.
Each family of clients gets its own endpoint:
- Claude.ai, Claude Code and Claude Cowork:
https://services.leadconnectorhq.com/mcp/anthropic/v2 - ChatGPT, Codex and other compatible OpenAI clients:
https://services.leadconnectorhq.com/mcp/openai/v2/ - Cursor:
https://services.leadconnectorhq.com/mcp/cursor/v2
An older universal endpoint, https://services.leadconnectorhq.com/mcp/, remains supported for any MCP client with a narrower set of operations, says a help article modified on 7 August 2026.
How does the MCP server sign in: OAuth or a private integration token?
The current guide uses OAuth for every client it covers. Adding the server to an assistant opens a LeadConnector sign-in page, where you choose the sub-account or sub-accounts to expose and approve the requested scopes. The assistant can then run only the operations those scopes allow, and access can be revoked from the LeadConnector account. The guide’s FAQ is direct about tokens: these connections need none, so there is nothing to create or rotate by hand.
Tokens remain an option. HighLevel’s article on multi-account support for Claude, modified on 6 August 2026, lists a private integration token as a second method: create it under Private Integrations in settings, select its scopes and send it as a Bearer token in the Authorization header. That article recommends OAuth because a token reaches fewer scopes, and the 7 August article says the universal endpoint accepts either method. HighLevel’s private integrations article sets the token rules: it is displayed once, rotation is advised every 90 days, and an agency or a sub-account can hold up to 20.
By default a connection is bound to one sub-account. Agency-wide connections that cover several sub-accounts are described as rolling out, and every request still runs against exactly one.
What can an AI assistant do through the MCP server?
Instead of hundreds of separate tools, the server exposes 4:
list_locationsreturns the sub-accounts the connection may use.search_operationsfinds the operations that match a task.describe_operationshows an operation’s inputs, required scopes and safety information.execute_operationruns it, within the scopes you granted.
Behind them sits the active catalog. The help article counts more than 600 read, create, update and delete operations across 38 domains; the developer page says more than 550. The article’s table lists 24 areas, among them:
- Contacts: search, create, update and delete, plus tags, notes, tasks and workflow enrollment.
- Conversations and messages: search conversations, read and send messages, recordings and transcriptions.
- Opportunities and pipelines: search, create, update, delete and status changes.
- Calendars and appointments: appointments, free and blocked slots, and bookings.
- Payments, invoices and store: orders, subscriptions, transactions, invoices, estimates and products.
- Workflows: read, create, update, publish and delete workflow definitions.
- Phone system and Voice AI: phone-number purchasing, agents and call logs.
The remaining areas include the social planner, blogs, email campaigns, forms, funnels, custom objects and the wallet. What one connection sees depends on its scopes, and HighLevel says sensitive and irreversible operations go through extra confirmation and safety checks. Two August 2026 articles give 625 operations across 40 domains, so expect the counts to move.
Which AI clients does HighLevel name, and how do they connect?
The help article gives setup steps for 8 clients:
- Claude.ai and Claude Cowork: add a custom connector in the connector settings, enter the Anthropic URL, then complete the LeadConnector sign-in.
- Claude Code: one
claude mcp addcommand with the same URL, or an entry in the project’s.mcp.jsonfile. - ChatGPT: the help article points to an official HighLevel plugin, which needs no server URL. The developer page instead describes turning on Developer mode and creating a plugin with the OpenAI URL.
- Codex:
codex mcp addwith the OpenAI URL, thencodex mcp login. - Cursor: an entry in
.cursor/mcp.jsonthat points at the Cursor URL. - Muse and Grok Bot: each lists HighLevel as an official app that connects through the same OAuth flow.
To confirm a connection, the guide suggests asking the assistant for the last 5 contacts added.
Which GoHighLevel plan does the MCP server need, and what does it cost?
Neither MCP page names a plan, and neither puts a price on the server. The guide’s stated requirements are access to the sub-accounts you want to connect and a supported AI client. HighLevel’s Pricing Guide, saved on 16 November 2026, lists no charge for API requests.
The nearest published rule is HighLevel’s API article, which sorts API access by plan:
- Starter, at $97 a month, and Unlimited, at $297 a month, get OAuth 2.0, private integration tokens and the standard endpoints, according to the article as rewritten on 13 November 2026.
- Founder, at $497 a month, is the plan the article ties its restricted endpoints to, naming two: creating a sub-account and enabling SaaS for one. It was called Agency Pro until HighLevel’s Founder Plan article of 10 November 2026, which keeps the price.
That rewrite removes an earlier conflict. The article used to file OAuth under advanced access, which sat oddly with an MCP server that signs in through OAuth; the current version lists OAuth on all three plans, and the MCP guide sets no plan condition. The article adds that an available sign-in method does not guarantee access to every endpoint, so test the operations you need before building a process on Starter or Unlimited. Our GoHighLevel pricing breakdown lists what each plan includes.
Three costs sit outside the server:
- The AI client. HighLevel’s documents quote no price for any of the 8 and say only that ChatGPT access depends on your ChatGPT plan.
- Usage. The catalog includes sending messages and buying phone numbers. HighLevel’s phone pricing bills a text at $0.00747 a segment before carrier fees and a local number at $1.15 a month, and the MCP pages mention no exemption.
- HighLevel’s own AI. The MCP server connects outside assistants. AI Employee and the pay-per-use AI features are priced separately, as our GoHighLevel AI pricing guide explains.
Does HighLevel publish rate limits for the MCP server?
Not for the server by name. HighLevel’s developer documentation has a rate limits page for its API, saved on 16 November 2026. It sets 100 requests per 10 seconds and 200,000 requests per 24 hours, counted for each Marketplace app on each sub-account or agency. Private integration tokens get the same defaults, but every token on one sub-account shares a single daily count.
Neither MCP page states whether an operation run through the server counts against those numbers. The multi-account article says each operation runs against the HighLevel API, so treat the API limits as the working assumption for bulk jobs.
How does the MCP server differ from the REST API and an n8n workflow?
All three reach the same CRM data. They differ in who decides what happens next.
- REST API. A developer writes each request to a documented endpoint and sends a token in the Authorization header. That suits fixed integrations inside other software, and HighLevel’s support team does not build or debug the code.
- MCP server. A person types a request, and the assistant finds the operations, checks their inputs and runs them. HighLevel’s own example is a single prompt that finds a contact, adds a tag and creates an opportunity.
- n8n workflow. A workflow repeats the same steps each time its trigger fires and can pass GoHighLevel data on to other apps. On the GoHighLevel side, premium webhook steps cost $0.01 a run.
The MCP server and n8n can also be combined. A HighLevel help article modified on 27 April 2026 says the server works with n8n without custom plugins, and its sample configuration uses the universal endpoint, a Bearer token and an optional location ID. Our GoHighLevel and n8n guide covers the webhook and API routes and their fees.
A workable split: keep unattended jobs that run on every lead in a workflow, and use the MCP server for questions and occasional changes a person reviews.
30 days free through HighLevel’s Bootcamp page, instead of the 14 on its pricing page. The signup form asks for payment details, usage fees for texts, calls and email can still be charged during the trial, and the plan bills monthly once it ends unless you cancel.
How we checked this
GoHighLevel: 104 vendor pages read, last verified . Prices are US list prices in USD.
Automated re-check on : all 36 prices we re-read were still on GoHighLevel's pages.
How GoHighLevel was checked: the method and 104 sources, and 17 points we could not verify
Read from HighLevel's pricing page (monthly, annual and Enterprise views), 18 help-center billing articles plus one CSV rate attachment, and the Terms of Service in a US-locale headless browser; snapshots saved.
- HighLevel pricing page (monthly view, add-ons, FAQ) — read 16 November 2026
- HighLevel pricing page (annual toggle, last shown on 26 September 2026) — read 26 September 2026
- HighLevel pricing page (Enterprise plan details) — read 16 November 2026
- HighLevel Pricing Guide (help center) — read 16 November 2026
- AI Product Pricing (help center) — read 16 November 2026
- Phone System Pricing & Billing Guide (help center) — read 16 November 2026
- Understanding the 5% Markup on Phone System Pass-Through Charges (help center) — read 16 November 2026
- A2P 10DLC Messaging Fees (help center) — read 16 November 2026
- What is LC Email? (pricing and sending limits, help center) — read 16 November 2026
- WhatsApp Pricing, Billing, and Rebilling Guide (help center) — read 16 November 2026
- WhatsApp Pricing - October 2026 CSV (attachment to the WhatsApp guide) — read 16 November 2026
- Rebilling, Reselling, and Wallets Explained (help center) — read 16 November 2026
- Activate SaaS Mode and Configure Phone Rebilling (help center) — read 16 November 2026
- Email Re-Billing on the Unlimited and Founder Plans (help center) — read 16 November 2026
- How to Enable and Rebill Workflow Premium Features (help center) — read 16 November 2026
- Workflows Pro Plan - New Pricing Tiers (help center) — read 16 November 2026
- What is a Dedicated IP? (help center) — read 16 November 2026
- Dedicated Email Sending Domains Overview & Setup (help center) — read 16 November 2026
- Billing Related Questions for Agencies (help center) — read 16 November 2026
- HighLevel API: access by plan level (help center) — read 16 November 2026
- Wallet Auto Recharge and Smart Adjustment Settings (help center) — read 16 November 2026
- Wallet Pricing Page (help center) — read 16 November 2026
- HighLevel Terms of Service (last updated June 2026) — read 16 November 2026
- HighLevel about us page — read 16 November 2026
- HighLevel affiliate agreement page — read 16 November 2026
- HighLevel affiliate brand kit page — read 16 November 2026
- HighLevel affiliate program policies page — read 16 November 2026
- HighLevel affiliate program page — read 16 November 2026
- HighLevel affiliate signup page — read 16 November 2026
- HighLevel ai page page — read 16 November 2026
- HighLevel bootcamp 30 day trial page — read 16 November 2026
- HighLevel crm page page — read 16 November 2026
- HighLevel developer docs: mcp server — read 16 November 2026
- HighLevel developer docs: rate limits — read 16 November 2026
- HighLevel help center: affiliate commander program — read 16 November 2026
- HighLevel help center: affiliate commission looks incorrect — read 16 November 2026
- HighLevel help center: affiliate first steps — read 16 November 2026
- HighLevel help center: affiliate integrishield compliance — read 16 November 2026
- HighLevel help center: affiliate payouts — read 16 November 2026
- HighLevel help center: affiliate portal — read 16 November 2026
- HighLevel help center: affiliate pro and trial links — read 16 November 2026
- HighLevel help center: affiliate referral statuses — read 16 November 2026
- HighLevel help center: affiliate split commissions — read 16 November 2026
- HighLevel help center: affiliate where is my payout — read 16 November 2026
- HighLevel help center: affiliates first steps — read 16 November 2026
- HighLevel help center: become an affiliate — read 16 November 2026
- HighLevel help center: founder plan — read 16 November 2026
- HighLevel help center: mcp anthropic scopes — read 16 November 2026
- HighLevel help center: mcp chatgpt codex — read 16 November 2026
- HighLevel help center: mcp claude multi account — read 16 November 2026
- HighLevel help center: mcp n8n — read 16 November 2026
- HighLevel help center: mcp server — read 16 November 2026
- HighLevel help center: premium support — read 16 November 2026
- HighLevel help center: saas configurator — read 16 November 2026
- HighLevel help center: saas mode faqs — read 16 November 2026
- HighLevel help center: saas multi brands — read 16 November 2026
- HighLevel help center: saas reselling add ons — read 16 November 2026
- HighLevel help center: saas v1 vs v2 — read 16 November 2026
- HighLevel help center: saas zero to sales — read 16 November 2026
- HighLevel help center: search free trial — read 16 November 2026
- HighLevel help center: trial implementation call — read 16 November 2026
- HighLevel help center: whitelabel desktop app setup — read 16 November 2026
- HighLevel help center: whitelabel domain — read 16 November 2026
- HighLevel help center: whitelabel mobile app — read 16 November 2026
- HighLevel help center: whitelabel mobile customizer — read 16 November 2026
- HighLevel home after click 1 page — read 16 November 2026
- HighLevel home after click 2 page — read 16 November 2026
- HighLevel home after click 3 page — read 16 November 2026
- HighLevel home after click 4 page — read 16 November 2026
- HighLevel home page — read 16 November 2026
- HighLevel pricing after click 1 page — read 16 November 2026
- HighLevel pricing annual check page — read 16 November 2026
- HighLevel pricing starter check page — read 16 November 2026
- HighLevel pricing page — read 16 November 2026
- HighLevel AI page (second capture) — read 16 November 2026
- HighLevel ecommerce page — read 16 November 2026
- HighLevel help center: API migration guide to private integrations — read 16 November 2026
- HighLevel help center: getting started with blogs — read 16 November 2026
- HighLevel help center: cancel location subscriptions from agency billing — read 16 November 2026
- HighLevel help center: what happens to your data when you cancel — read 16 November 2026
- HighLevel help center: connecting your domain — read 16 November 2026
- HighLevel help center: domain setup for funnels and websites — read 16 November 2026
- HighLevel help center: ecommerce store setup — read 16 November 2026
- HighLevel help center: export contacts as CSV — read 16 November 2026
- HighLevel help center: Funnel AI in Ask AI — read 16 November 2026
- HighLevel help center: Funnel and Website AI — read 16 November 2026
- HighLevel help center: how to launch a funnel — read 16 November 2026
- HighLevel help center: Pause Account feature ($27 Pause Plan) — read 16 November 2026
- HighLevel help center: releasing purchased domains from a canceled account — read 16 November 2026
- HighLevel help center: restore a deleted phone number — read 16 November 2026
- HighLevel help center: search zapier — read 16 November 2026
- HighLevel help center: template library — read 16 November 2026
- HighLevel help center: launch a website from a template — read 16 November 2026
- HighLevel help center: websites overview — read 16 November 2026
- HighLevel help center: WordPress hosting plans — read 16 November 2026
- HighLevel help center: WordPress hosting setup guide — read 16 November 2026
- HighLevel help center: connect a sub-account with Zapier — read 16 November 2026
- HighLevel help center: create sub-accounts using Zapier — read 16 November 2026
- HighLevel help center: Zapier duplicate contacts — read 16 November 2026
- HighLevel help center: Zapier troubleshooting — read 16 November 2026
- HighLevel help center: webhooks with Zapier — read 16 November 2026
- HighLevel pricing page (annual prices read from hidden page elements; toggle not shown) — read 16 November 2026
- HighLevel website builder page — read 16 November 2026
- HighLevel WordPress page — read 16 November 2026
- Whether the 14-day trial started from the pricing page requires a card up front: that signup form asks for business details first and its 'Trial Details' step was not reached, on 26 September or on 16 November 2026. The pricing page says billing starts automatically when the trial ends unless canceled. This applies only to the 14-day route: the 30-day trial on HighLevel's Bootcamp page shows a Payment Information step in its signup form (snapshots of 6 October and 16 November 2026).
- The 5% pass-through markup on A2P and carrier fees: the markup article (9 Sep 2026) says it applies whether or not rebilling is enabled, while the Phone System guide, as revised on 12 October 2026, says it applies only when re-billing is enabled, and the A2P fee article says HighLevel adds no markup. Both articles were read again on 16 November 2026 and still disagree, and the rebilling articles re-read that day do not settle it. Starter cannot rebill at all, so on the guide's reading the scenarios would not carry it. The scenarios include it; without it they fall by $0.12 (solo), $0.30 (small) and $1.20 (growing).
- The Phone System guide lists A2P monthly fees with the 5% built in (Sole Proprietor 'up to $2.10', Low Volume '$1.50-$10.50', High Volume '$10.50'), while the A2P article lists $2.00 / $1.50 / $10.00. The scenarios use the A2P article's $1.50.
- Carrier fees depend on each recipient's carrier; the scenarios assume $0.0045 per outbound segment for every text, which is T-Mobile's rate. Verizon and US Cellular are $0.0050, AT&T $0.0035 and other carriers $0.0042 as of 16 November 2026.
- Pay-per-use Conversation AI and Voice AI cost per conversation depends on token counts and cannot be priced in advance; only the per-token and per-minute rates are recorded.
- The per-message price for US WhatsApp service messages beyond the free 1,000 a month (in force since 1 October 2026) appears only in an attached CSV ($0.0034 before the 5% fee), not in the guide's table, whose Service column still shows the rates through 30 September. The CSV, downloaded again on 16 November 2026, is unchanged ($0.0250 marketing, $0.0034 utility and service for North America). The guide's US utility rate now equals the CSV's $0.0034, so whether the 5% fee is added on top of it is not stated.
- HighLevel's pricing pages list no platform fee on payments taken through Stripe, PayPal, Square and other connected processors, but processor fees were not captured.
- Domain purchase prices vary by TLD and were not captured.
- Whether a dedicated email sending domain costs anything: the setup article lists no fee, but does not say it is free.
- The $27/month Pause Plan: HighLevel's Pause Account article describes it for agency users and its billing FAQ quotes the same price, but the same FAQ also says agency accounts cannot be paused or placed on hold.
- Enterprise pricing is quote-only.
- HighLevel's in-app Wallet Pricing page (the source it calls authoritative for usage rates) requires a login and was not captured.
- The dedicated email IP's unit: the pricing page and the dedicated IP article say $59 a month per IP, while HighLevel's Pricing Guide says "$59/mo per dedicated domain". The record follows the two that agree.
- The Workflow Pro unit: HighLevel's Pricing Guide calls the $10 tier a price "for the agency" in one line but lists it per enabled sub-account in its table, and the Workflow Pro article counts allowances per sub-account. The record treats it as per sub-account.
- Annual plan prices: the pricing page as saved on 16 November 2026 shows no annual toggle and no yearly prices. $970, $2,970 and $4,970 remain in page elements the stylesheet hides (the $2,970 and $4,970 elements sit in the Founder and Unlimited cards the wrong way round), and the dedicated IP article (5 November 2026) still names the $297/$2970 and $497/$4970 plans, as do the phone and email rebilling articles (both modified 27 October 2026). Starter's $970 has no visible source newer than 26 September 2026. The record keeps the September annual prices; confirm at checkout.
- Plan card rewording on the pricing page (16 November 2026): Starter's card no longer lists '3 Sub-Accounts' or '24/7 Support' (the page's FAQ still gives 3 sub-accounts, and the card now says 'Chat and phone support'); 'White-label desktop app' now sits on the Unlimited card, while the page's FAQ says without naming a plan that you can white-label the desktop web app, so whether Starter includes it is not stated; the Founder card no longer lists user/agent reporting or advanced API access (HighLevel's Founder article confirms advanced API access).
- API access by plan: HighLevel's API article, rewritten on 13 November 2026, no longer separates location-level keys from agency-level keys and OAuth. It gives Starter, Unlimited and Founder OAuth 2.0 and Private Integration Tokens, documents the Create Sub-Account and Enable SaaS endpoints as needing Agency Pro, and says those restrictions still have to be mapped to the Founder plan. The pricing page lists 'Basic API access' on the Unlimited card only, and the Founder article says advanced API access is included. Which other endpoints are restricted is not stated.
We re-check every price against the vendor's page each Monday; changes are logged on the changes page. See a price that's wrong? Tell us and we will correct it the same week.
Frequently asked questions
What is the GoHighLevel MCP server?
The GoHighLevel MCP server is a connection point that lets an AI assistant work inside a GoHighLevel sub-account. MCP is short for Model Context Protocol, a standard way for assistants to discover and use an application's tools, and HighLevel also calls its version the LeadConnector MCP Server. Its help article, last modified on 15 November 2026, says the server connects through OAuth: you sign in, choose the sub-account or sub-accounts to include and approve the requested permissions. After that the assistant can search records and run supported read, create, update and delete operations. The article counts more than 600 operations across 38 domains, including contacts, conversations, opportunities, calendars, payments and workflows, all reached through 4 tools. HighLevel's developer documentation carries the same material under the title LeadConnector MCP Server.
How do I connect Claude to GoHighLevel with MCP?
HighLevel's setup guide gives Claude its own endpoint, https://services.leadconnectorhq.com/mcp/anthropic/v2, which serves Claude.ai, Claude Code and Claude Cowork. In Claude.ai, the documented route is to add a custom connector under Settings and Connectors, enter that URL, click Connect and finish the LeadConnector sign-in by choosing a sub-account and approving the requested access. A new chat then has the LeadConnector tools. In Claude Code, the guide gives a single claude mcp add command with the same URL, or an entry in the project's .mcp.json file, and a browser window opens for authorization on first use. No private integration token is needed for these connections. To check the result, HighLevel suggests asking Claude to find the last 5 contacts added in HighLevel. Agencies can include several sub-accounts where the multi-sub-account install option has been enabled.
Does the GoHighLevel MCP server cost extra?
HighLevel publishes no price for the MCP server. Its setup guide and its developer documentation, both saved on 16 November 2026, list no subscription, add-on or per-operation fee, and the only stated requirements are access to the sub-accounts you connect and a supported AI client. Costs can still arise around it. GoHighLevel plans start at $97 a month for Starter, with Unlimited at $297 and Agency Pro at $497. Actions an assistant takes are not exempted from usage pricing in those documents, so a text it sends would bill at the normal rate of $0.00747 a segment before carrier fees. The AI assistant is a separate product that HighLevel does not price, and HighLevel's own AI features, such as AI Employee at $50 or $97 per sub-account a month, are charged separately.
Which GoHighLevel plan do I need for the MCP server?
HighLevel's MCP documentation names no plan. The setup guide asks only for access to the sub-account you want to connect and a supported AI client. The closest published rule is HighLevel's API article, rewritten on 13 November 2026, which lists OAuth 2.0 and private integration tokens as available on Starter at $97 a month, Unlimited at $297 and Founder at $497. The MCP server signs in through OAuth, so nothing in that article keeps it from the two lower plans. The article does reserve some endpoints for the top plan, naming sub-account creation and SaaS enablement, and warns that having a sign-in method does not guarantee access to every endpoint. Test the operations you need before relying on the server on a lower plan. Agency-wide connections that span several sub-accounts are described as still rolling out.
Does the GoHighLevel MCP server need a private integration token?
Not for the connections in HighLevel's current setup guide. Claude, ChatGPT, Codex, Cursor, Muse and Grok Bot all connect through OAuth, and the guide's own FAQ says there is no token to create or rotate. A token remains an option on the older universal endpoint, https://services.leadconnectorhq.com/mcp/. HighLevel's article on multi-account support for Claude, modified on 6 August 2026, describes creating a private integration token in settings, selecting its scopes and sending it as a Bearer token in the Authorization header. That article recommends OAuth because a token reaches a more limited set of scopes. If you do use a token, HighLevel's private integrations article says it is shown only once and should be rotated every 90 days, and that an agency or a sub-account can hold up to 20 of them.
What is the difference between the GoHighLevel MCP server and the GoHighLevel API?
The API is for code, and the MCP server is for AI assistants. With the REST API, a developer sends each request to a documented endpoint and authenticates with a private integration token or OAuth; HighLevel's API article says its support team does not help build or debug those integrations. With the MCP server, a person types a request and the assistant uses 4 tools to search for the right operation, inspect its inputs and run it, so nobody writes requests by hand. HighLevel says each MCP operation runs against its API and stays within the scopes approved at sign-in. The API article publishes limits of 100 requests per 10 seconds and 200,000 a day for each app on each sub-account. The MCP pages publish no separate limit and do not say whether assistant activity counts toward those numbers.
First published 16 November 2026 · Last updated