Best MCP Servers for ChatGPT (2026)
Last updated 2026-10-09
Quick answer
The best MCP servers for ChatGPT are remote, hosted servers: GitHub, Context7, Figma, Supabase, Notion, Linear, Cloudflare, Stripe and Sentry are the ones most worth connecting in 2026. ChatGPT reaches MCP through connectors (Settings → Connectors) and, for full tool calls, developer mode — both favor Streamable HTTP endpoints over locally installed servers.
How ChatGPT connects to MCP (different from Claude or Cursor)
Claude Desktop and Cursor run local servers over stdio. ChatGPT is cloud-first: it connects to remote MCP servers over Streamable HTTP, either as connectors that supply search-and-retrieve context, or — with developer mode enabled — as full tool servers the model can call directly. That means the practical answer to 'which MCP servers work with ChatGPT' is: the ones that publish a hosted HTTP endpoint.
Most vendors now offer one. Figma, Linear, Stripe, Cloudflare and Atlassian all publish official remote endpoints; community hubs list hundreds more. Directory entries on MCPNav show a Remote (Streamable HTTP) badge and a URL-based config for every server that supports it.
Top remote MCP servers worth connecting to ChatGPT
Ranked by how much capability they add to ChatGPT in daily work:
- ▸GitHub MCP — repositories, issues, pull requests (free / open source, remote options available)
- ▸Context7 — current library and framework documentation (free tier)
- ▸Figma MCP (official remote) — design files and UI context (plan dependent)
- ▸Supabase MCP — databases and project management (free tier)
- ▸Notion MCP — documents, wikis and workspace data (plan dependent)
- ▸Linear MCP (official remote) — issues, projects, workflows (plan dependent)
- ▸Cloudflare MCP — infrastructure and operational tools (free)
- ▸Stripe MCP — payments and business data (free, per-action approval)
- ▸Sentry MCP — production errors and debugging (free tier)
- ▸Jira MCP — issues and Confluence context via hosted endpoints (plan dependent)
What ChatGPT can and cannot do with MCP today
Connectors shine at retrieval: point ChatGPT at a knowledge base and it can search and cite it inside answers and deep research. Developer mode is where servers become tools — creating issues, updating records, triggering deploys. Two practical limits: only remote (HTTP) endpoints work — a stdio-only server like the reference Filesystem server has no ChatGPT path — and admin settings govern what workspace members can connect, so personal plans have fewer options than team plans.
If a server you want is local-only, check whether the publisher also offers a hosted endpoint, or run a bridge such as mcp-remote that exposes a local server over HTTP. Verify any bridge against your security requirements before connecting it to a cloud client.
Safety notes
Connecting a remote server hands a cloud provider an authorised token to that system. Scope it: read-only first, per-project credentials where supported, and revoke what you are not using. Stripe's server is the good pattern here — consequential actions require explicit approval. Review each server's tool list before connecting; a directory listing is discovery, not a security audit. More: MCP security best practices.
The picks, one by one
The official GitHub server, reachable remotely via hosted endpoints — repo search, issues and PRs inside ChatGPT research.
Best for: Deep research over your own repositories and engineering work.
Watch out: Scope tokens read-only unless you enable developer mode tool calls.
Current, version-specific library documentation served on demand.
Best for: Grounding answers in docs newer than the model's training data.
Watch out: Free tier limits apply.
Figma's official remote endpoint — design file context without local installs.
Best for: Frontend teams pulling real design specs into implementation chat.
Watch out: Needs a Figma plan that includes MCP.
Database schema, queries and project management for Supabase backends.
Best for: Full-stack work where ChatGPT needs real schema context.
Watch out: Point it at a dev project, not production.
Official Notion server: search and read pages and databases (remote options available).
Best for: Answering from your team wiki inside ChatGPT.
Watch out: Wide scopes — audit what the integration can read.
Linear's official remote MCP server: issues, projects and workflows over Streamable HTTP.
Best for: Project status questions and triage straight from chat.
Watch out: Plan dependent; start with read scopes.
Official Cloudflare server for infrastructure inspection and operations.
Best for: Site and platform operators managing DNS, workers and zones.
Watch out: Powerful credentials — use scoped API tokens.
Hosted Stripe server with per-action approval for payments operations.
Best for: Billing and invoice questions grounded in live data.
Watch out: Use restricted keys; test mode first.
Official Sentry server: production errors and traces into the research loop.
Best for: Debugging with real telemetry instead of pasted logs.
Watch out: Read-only by default — keep it that way.
Hosted Jira endpoint for issues and project context (Streamable HTTP/SSE).
Best for: Sprint and ticket answers without leaving the chat.
Watch out: Third-party hosting — verify the operator before granting tokens.
Frequently asked questions
Does ChatGPT support MCP servers?＋
Yes, through connectors and developer mode. Connectors let ChatGPT search and retrieve from a connected service; developer mode (on supported plans) enables full tool calls to remote MCP servers over Streamable HTTP.
Can ChatGPT use local stdio MCP servers?＋
No — ChatGPT connects to remote HTTP endpoints only. Local-only servers need a publisher-hosted endpoint or a self-hosted bridge that exposes them over HTTP, which you should vet carefully.
How do I add an MCP server to ChatGPT?＋
Settings → Connectors → Create/Connect, then paste the server's remote MCP URL (and auth). For tool calls, enable developer mode in Settings → Apps & Connectors and approve the server for your workspace.
Which ChatGPT plans include MCP connectors?＋
Connectors and deep research integrations are available on paid plans, with administrator controls on team and enterprise workspaces. Feature availability moves fast — check OpenAI's current docs for your plan.