This list is part of our best-of MCP lists, within The MCP Guide. Entries are checked against each vendor's own pages; inclusion is editorial and not paid.
The best MCP servers for ChatGPT in 2026 are official servers from HubSpot, Box, Atlassian, Asana, Airtable, Stripe, Zapier, Figma, Linear and Vercel. Each vendor documents how to connect ChatGPT, and most use OAuth so you sign in with your own account.
ChatGPT now packages MCP connections as plugins. Many vendors are already in ChatGPT's plugins directory. For others, OpenAI's custom MCP server guide says you open Plugins, choose "Add custom MCP server," and then "Create as a plugin."
We checked each server against the vendor's own docs on 28 October 2026. eSEOspace did not test these servers.
Key Takeaways
- In ChatGPT on the web, MCP servers reach you as plugins; the ChatGPT desktop app, Codex CLI and Codex IDE extension can also connect to MCP servers directly and share one configuration.
- Adding your own custom MCP server is a Business, Enterprise and Edu feature on the web as of October 2026; Pro users get read and fetch only.
- OpenAI says "ChatGPT currently requires manual confirmation in any conversation before write actions can be taken."
- OpenAI also warns that custom MCP servers "are not developed or verified by OpenAI," so stick to official vendor servers.
- Most vendors list no separate MCP price, but some count calls against your plan, such as Zapier tasks or Box API calls.
How we chose
We included a server only if it passed these tests:
- Official. Published by the product's own company.
- Documented for ChatGPT. The vendor's docs name ChatGPT or give ChatGPT setup steps, and we opened that page on 28 October 2026.
- Clear access model. The docs explain sign-in, permissions or admin controls.
- Clear cost. The docs state plan rules or say there is no extra charge.
We left out eSEOspace and our clients. OpenAI's help center blocked our automated page check on 28 October 2026, so plan rules for custom servers come from our earlier check of that page and from OpenAI's developer docs, both linked below. For the bigger picture, see our best MCP clients list.
Comparison table
| Name | Best for | Pricing | Notable |
|---|---|---|---|
| HubSpot | CRM, deals and marketing content | All HubSpot plans | Reads and writes, never deletes |
| Box | Files with strict permissions | All Box plans; some calls metered | Admin enables it in Box first |
| Atlassian Rovo MCP | Jira and Confluence | Uses Rovo credits per call | OAuth 2.1, many Atlassian products |
| Asana | Tasks and projects | No plan requirement stated | Official Asana app in ChatGPT |
| Airtable | Bases, records and forms | All plans, no extra charge | 10 records per create request |
| Stripe | Payments, invoices and subscriptions | No separate MCP price listed | Refunds need human confirmation |
| Zapier | Actions in thousands of apps | 2 Zapier tasks per tool call | Business, Enterprise or Edu workspace app |
| Figma | FigJam diagrams and Slides decks | Any Figma plan | Not available in the EU |
| Linear | Issue context in deep research | No separate MCP price listed | Deep research on Plus and Pro too |
| Vercel | Deploys, logs and analytics | All Vercel plans | Only approved clients can connect |
1. HubSpot: CRM records in ChatGPT
What it is. HubSpot's ChatGPT guide says the connector reads and updates contacts, companies, deals, tickets, leads, campaigns, blog posts and more. Quotes, invoices, orders and analytics are read-only.
It cannot delete any object. HubSpot recommends you "configure the connector to always ask for approval before updates are made."
Best for: Sales and marketing teams that want CRM answers and updates from chat.
Pricing: Available with "All products and plans" on the HubSpot side. Our HubSpot connect guide covers setup.
Pros
- Works on every HubSpot plan.
- No delete actions.
- Covers marketing data, not just CRM.
Cons
- Cannot read Sensitive Data properties.
- If Sensitive Data is on, engagement data is blocked entirely.
Checked against: HubSpot ChatGPT connector guide on 28 October 2026.
2. Box: governed file access
What it is. Box's ChatGPT setup page says a Box admin first sets the ChatGPT MCP Server integration to "Available to all users." Users then open Apps in ChatGPT, search for Box, and connect.
Enterprise customers can ask OpenAI to limit connections to their verified company domain.
Best for: Legal, finance and HR teams that keep sensitive files in Box.
Pricing: Box's MCP pricing page says the server is "available on all Box plans." Calls are free only with a published app and your own OAuth login; other cases are charged per API call, and AI tools also use AI Units. See our Box connect guide.
Pros
- Admin must turn it on, which suits regulated teams.
- Domain restriction option on Enterprise.
- Available on every Box plan.
Cons
- Metered calls outside the standard setup.
- AI tools need a Box AI-enabled plan.
Checked against: Box ChatGPT setup docs on 28 October 2026.
3. Atlassian Rovo MCP Server: Jira and Confluence
What it is. Atlassian's getting started page lists OpenAI ChatGPT as a supported client. It can "Summarize and search Jira, Jira Service Management, Confluence, Bitbucket, Projects, Goals, and more."
It signs in with OAuth 2.1, and API tokens are optional. It acts with each user's existing permissions.
Best for: Teams that plan in Jira and document in Confluence.
Pricing: Calls use Rovo credits from your organization's shared pool, and "credit usage is per call." See our Atlassian connect guide.
Pros
- Broad coverage of Atlassian cloud products.
- Existing permissions carry over.
- Works in many clients, not just ChatGPT.
Cons
- Each call uses Rovo credits.
- v1 usage moves to v2 tools on 1 March 2027.
Checked against: Atlassian Rovo MCP getting started on 28 October 2026.
4. Asana: tasks and projects
What it is. Asana's MCP docs list "the Asana app in ChatGPT" alongside Claude, Codex, Cursor and VS Code. The current V2 server lives at https://mcp.asana.com/v2/mcp.
The beta V1 endpoint is deprecated and being shut down, so older setups need to move.
Best for: Teams that track marketing or operations work in Asana.
Pricing: No required plan is named. Enterprise+ and Legacy Enterprise customers can allow or block MCP clients.
Pros
- Official Asana app in ChatGPT.
- OAuth with your own account.
- Admin client controls on top tiers.
Cons
- V1 users must migrate.
- Tool list is only visible from inside a client.
Checked against: Asana MCP docs on 28 October 2026.
5. Airtable: bases, records and forms
What it is. Airtable's MCP guide names "Claude, ChatGPT, and other MCP-compatible tools" and has a ChatGPT quickstart. It can search data, create and update records, build bases and forms, and draft automations.
"Creating records is limited to 10 records per request." It cannot deploy automations or edit existing interfaces.
Best for: Operations teams that run trackers in Airtable.
Pricing: "All plans." The guide says MCP is included at no extra charge, but API rate limits and base limits still apply. See our Airtable connect guide.
Pros
- No extra charge on any plan.
- Read-only scope option.
- Enterprise admins can allowlist the integration.
Cons
- 10-record cap per create request.
- Cannot deploy automations.
Checked against: Airtable MCP guide on 28 October 2026.
6. Stripe: payments and billing
What it is. Stripe's MCP docs link to an official Stripe plugin in ChatGPT that signs in with OAuth. Its tools read and write through the Stripe API, covering customers, invoices, subscriptions, refunds and more.
Stripe "requires human confirmation before it takes certain stripe_api_write actions, such as refunds and outbound payments." From December 2026 the server stops accepting full-access secret keys and restricted keys without the Agent tag.
Best for: Finance and support teams that look up payments or draft invoices.
Pricing: The docs list no separate MCP price; normal Stripe fees apply to anything the agent does. See our Stripe connect guide.
Pros
- Built-in confirmation for risky writes.
- Admins can turn MCP access off for live or sandbox.
- OAuth sessions can be revoked per user.
Cons
- Broad write access if you approve it.
- API key users must migrate by 30 November 2026.
Checked against: Stripe MCP docs on 28 October 2026.
7. Zapier: actions across thousands of apps
What it is. Zapier's ChatGPT guide says you need "Business, Enterprise, or Edu" if your workspace already added Zapier. Otherwise, it describes a developer mode route that creates a private plugin pointing at https://mcp.zapier.com/api/v1/connect.
A plugin you create this way is private to your account until a workspace Owner publishes it.
Best for: Teams that need ChatGPT to act in apps with no official server.
Pricing: "Each successful tool call uses two tasks from your Zapier plan." See our Zapier connect guide.
Pros
- Reaches apps that lack their own server.
- Owner controls who gets the plugin.
- One sign-in for many apps.
Cons
- Every call uses Zapier tasks.
- Zapier's guide still mentions developer mode, which may not match your ChatGPT plan's menus.
Checked against: Zapier ChatGPT guide on 28 October 2026.
8. Figma: diagrams and slides from chat
What it is. Figma's help article says "You can create and open Figma files directly from ChatGPT." It generates FigJam diagrams, such as flowcharts, and Figma Slides presentations.
It only creates FigJam and Slides files, not design files. The article says "This feature is currently not available in the EU."
Best for: Teams that want quick diagrams or first-draft decks.
Pricing: "Available on any Figma plan" for FigJam and Figma Slides. The Slides integration is in beta and may have usage limits. See our Figma connect guide.
Pros
- Works on any Figma plan.
- Output opens in Figma for editing.
- Good for diagrams from plain text.
Cons
- Not available in the EU.
- Does not create full Figma design files.
Checked against: Figma app in ChatGPT help on 28 October 2026.
9. Linear: issue context for deep research
What it is. Linear's ChatGPT page says the connector lets deep research "surface Linear issues, projects, and discussions relevant to your current context." It keeps "existing user-level permissions."
Linear also runs a general MCP server with a separate read-only endpoint for other clients.
Best for: Product teams that want research reports grounded in their own issues.
Pricing: No separate price. Linear says the connector "is available in deep research for Plus & Pro users (excl. EEA, CH, UK)," and on Team, Enterprise and Edu. See our Linear connect guide.
Pros
- Available beyond Business and Enterprise.
- Respects user permissions.
- Read-only endpoint exists for other clients.
Cons
- Focused on deep research, not issue editing.
- Plus and Pro access excludes the EEA, Switzerland and the UK.
Checked against: Linear ChatGPT integration page on 28 October 2026.
10. Vercel: deploys and logs
What it is. Vercel's MCP docs list ChatGPT as a supported client. You open Plugins, select "Add custom MCP server," enter https://mcp.vercel.com, pick OAuth, and select "Create as a plugin."
Vercel "only supports AI clients that have been reviewed and approved by Vercel." It can search docs, manage projects and deployments, read logs and query analytics.
Best for: Web teams that host on Vercel and want deploy status in chat.
Pricing: "Vercel MCP is available on all plans," subject to Vercel's AI Product Terms. More in our Vercel MCP server profile.
Pros
- Approved-client list blocks unknown apps.
- Explicit consent for each client connection.
- Available on all Vercel plans.
Cons
- Custom-server route needs a plan that allows it.
- Grants the same access as your Vercel account.
Checked against: Vercel MCP docs on 28 October 2026.
Plans and safety in ChatGPT
Our earlier check of OpenAI's help center found custom MCP servers on Business, Enterprise and Edu on the web, with Pro limited to read and fetch. Plus and Free were not listed. Directory plugins and deep research connectors such as Linear can reach more plans, so check each vendor's page.
On the desktop app, OpenAI's MCP page says each server can have an approval mode: auto, prompt, writes or approve. The writes mode "prompts for tools that aren't marked read-only." That is a sensible default for anything that can change data.
Next step
If you publish an MCP server and want it found by ChatGPT users, see our MCP marketing services. To compare more options, browse the MCP Guide's best-of section.
Frequently asked questions
Can I add any MCP server to ChatGPT?
On the web, adding a custom MCP server is available on Business, Enterprise and Edu as of October 2026, with Pro limited to read and fetch. You open Plugins, choose "Add custom MCP server," and create it as a plugin. Workspace admins can control which plugins are available.
What is the difference between a ChatGPT app, plugin and connector?
They overlap. OpenAI's developer docs say "A plugin packages an MCP connection for use in ChatGPT." Vendors still use the words app and connector in their own guides, but all of them point to an MCP server underneath.
Will ChatGPT change my data without asking?
OpenAI says ChatGPT requires manual confirmation before write actions in a conversation. Some vendors add their own checks too, such as Stripe's human confirmation for refunds. Still, only approve writes you understand.
Does ChatGPT desktop support local MCP servers?
Yes, according to OpenAI's MCP page. The desktop app can add a STDIO or Streamable HTTP server under Settings, and it shares its MCP configuration with the Codex CLI and IDE extension.
Are custom MCP servers safe to connect?
Only if you trust the company behind them. OpenAI warns that custom MCP servers "are not developed or verified by OpenAI." Prefer the official vendor servers on this list.
