An MCP server is a program that exposes tools, resources and prompts to AI applications through the Model Context Protocol, and the hard part for most teams is no longer finding one but knowing which servers are official, how they connect and what access they ask for. This hub lists the reference servers maintained by the MCP project and a set of widely used first-party servers from Atlassian, Cloudflare, Context7, GitHub, Notion, Playwright, Sentry and Stripe, with each one's endpoint, local option, authentication methods, repository licence and current status. It also explains how the official MCP Registry works and how to bring servers into an organisation in a controlled way. It is written for developers, platform teams and the people who approve which servers may be connected. Every fact comes from the project's or vendor's own repository or documentation. Last checked: 2026-11-15.
Servers are listed alphabetically; order is never a ranking.
This is an independent reference page from NHI Mgmt Group based on public provider documentation; it is not sponsored or endorsed by any provider.
Key takeaways
- The reference servers are teaching code. The MCP project's servers repository holds seven active reference servers and says they are educational examples, "not as production-ready solutions".
- The Registry is the official catalogue. The MCP Registry at registry.modelcontextprotocol.io launched in preview on 8 September 2025, and its README recorded an API freeze at v0.1 from 24 October 2025.
- Most major servers are now remote. Atlassian, Cloudflare, Context7, GitHub, Notion, Sentry and Stripe each publish a hosted Streamable HTTP endpoint; Playwright MCP runs locally over stdio by default.
- OAuth is the common front door. Every remote server in this hub supports OAuth or a vendor sign-in flow, and most also accept a scoped token for clients or automation that cannot use OAuth.
- Endpoints and tools change. Atlassian moved to a v2 URL, Cloudflare retired its
/sseURLs, Notion version 2.0 renamed tools and Stripe changes which API keys it accepts after 30 November 2026, so pin versions and review changes.
What an MCP server is, and what counts as official here
An MCP client (an IDE, a chat application or an agent) connects to one or more servers. A local server runs on the user's machine and usually talks to the client over stdio. A remote server is hosted by its operator and reached over Streamable HTTP, typically with OAuth. For the protocol itself, see What Is MCP? The Model Context Protocol Explained (2026); for the transports, see Streamable HTTP vs stdio MCP Transport.
In this hub, official means one of two things: a reference server maintained by the MCP steering group in the modelcontextprotocol/servers repository, or a server published by the company whose product it connects to, in that company's own repository or documentation. Community servers that wrap a vendor's API are out of scope, because their names alone do not show who maintains them.
How to find servers: the official MCP Registry
The servers repository now sends readers who want a list of servers to the MCP Registry. The announcement post describes the Registry as "an open catalog and API for publicly available MCP servers", and the open source registry project describes itself as "like an app store for MCP servers".
| Aspect | What the MCP project states |
|---|---|
| Official instance | registry.modelcontextprotocol.io, with live API documentation at /docs |
| Status | Preview since 8 September 2025; no data durability guarantees, and breaking changes or data resets may occur before general availability |
| API stability | API freeze at v0.1 announced in the README on 24 October 2025 |
| Metadata | Each server is described in a server.json file |
| Namespaces | GitHub-based publishing requires names of the form io.github.username/*; domain-based publishing requires the reverse-DNS form of a domain the publisher controls, such as com.example.*/* |
| Subregistries | Public subregistries (for example client marketplaces) can add to the upstream data; private subregistries can run inside enterprises using the Registry as their upstream |
| Moderation | Community members can flag spam, malicious code or impersonation; maintainers can denylist entries and remove them from public access |
Table as of 15 November 2026. A namespace proves who published an entry, not that the server is safe to connect. The security side of publishing and consuming registry entries is covered in the MCP Registry and Server Publishing Security Guide, and the choice between a registry and a gateway in MCP Gateway vs MCP Registry.
Reference servers from the MCP project
The steering group maintains a small set of servers to demonstrate protocol features and the official SDKs. TypeScript servers start with npx and Python servers with uvx or pip.
| Server | What it demonstrates, as described in the repository |
|---|---|
| Everything | Reference and test server with prompts, resources and tools |
| Fetch | Web content fetching and conversion for efficient LLM use |
| Filesystem | File operations with configurable access controls |
| Git | Tools to read, search and manipulate Git repositories |
| Memory | Knowledge graph based persistent memory |
| Sequential Thinking | Reflective problem solving through thought sequences |
| Time | Time and time zone conversion |
Table as of 15 November 2026. Thirteen earlier reference servers, including GitHub, GitLab, PostgreSQL, Puppeteer, Redis, Sentry, Slack and SQLite, have moved to the servers-archived repository. For several the repository points to a maintained successor (for example Brave's own Brave Search server, and a Slack server now maintained by Zencoder); GitHub and Sentry now publish their own servers, listed below. If a configuration still launches an archived package such as the old GitHub or Sentry reference server, replace it. The repository's licence file records a transition from MIT to Apache-2.0 for code, with documentation under CC-BY-4.0.
Major first-party servers at a glance
| Server | Publisher | Remote endpoint | Local option | Authentication | Repository licence |
|---|---|---|---|---|---|
| Atlassian Rovo MCP Server | Atlassian | https://mcp.atlassian.com/v2/mcp | Local clients via the mcp-remote proxy | OAuth 2.1; optional API token | Apache-2.0 |
| Cloudflare MCP servers | Cloudflare | https://mcp.cloudflare.com/mcp plus product servers such as docs.mcp.cloudflare.com/mcp | Servers are published as remote endpoints | OAuth; Cloudflare API token with the scopes each server needs | Apache-2.0 |
| Context7 | Upstash | https://mcp.context7.com/mcp | CLI and skills mode as an alternative to MCP | OAuth during setup; API key as a bearer token | MIT |
| GitHub MCP Server | GitHub | https://api.githubcopilot.com/mcp/ | Local binary or Docker image ghcr.io/github/github-mcp-server | OAuth; personal access token | MIT |
| Notion MCP | Notion | https://mcp.notion.com/mcp (SSE fallback at /sse) | mcp-remote for stdio-only clients; open source local server | OAuth; MCP tokens; enterprise-managed authorisation | MIT (local server) |
| Playwright MCP | Microsoft | Runs where you start it; SSE port option | npx @playwright/mcp@latest over stdio | Not applicable: runs a local browser | Apache-2.0 |
| Sentry MCP | Sentry | https://mcp.sentry.dev/mcp | sentry mcp in the Sentry CLI over stdio | OAuth; Sentry user auth token for stdio | FSL-1.1-Apache-2.0 |
| Stripe MCP | Stripe | https://mcp.stripe.com | Configured through Stripe's agent plugins and CLI | OAuth; agent API keys | MIT (agent toolkit) |
Table as of 15 November 2026. Repository licence applies to the published code; use of a hosted endpoint is governed by the provider's own terms.
Servers by category
| Category | Servers in this hub | Typical uses stated by the provider |
|---|---|---|
| Browser automation | Playwright MCP; Cloudflare Browser Run server | Page interaction through accessibility snapshots; fetching pages, converting to markdown, screenshots |
| Cloud and infrastructure | Cloudflare MCP servers | Broad Cloudflare API access through code execution; containers, monitoring and CASB servers |
| Code hosting and developer platforms | GitHub MCP Server; Git reference server | Repositories, issues, pull requests, Actions, code security |
| Documentation and knowledge | Context7; Cloudflare documentation server; Fetch reference server | Version-specific library documentation and code examples in the agent's context |
| Observability and errors | Sentry MCP | Searching errors, analysing performance, triaging issues |
| Payments and finance | Stripe MCP | Reading and writing Stripe API objects, documentation search, analytics |
| Work management and content | Atlassian Rovo MCP Server; Notion MCP | Jira, Confluence, Jira Service Management, Bitbucket, Loom; Notion pages, databases and search |
Security tooling servers (SIEM, EDR, cloud posture and similar) are covered separately in the MCP Servers for Security Tools Guide.
Server notes: what each publisher documents
Atlassian Rovo MCP Server
A cloud-hosted server for Atlassian Cloud covering Jira, Jira Service Management, Confluence, Bitbucket, Compass, Loom and wider platform features. The README says v2 is the recommended version, that new setups should use the /v2/mcp URL, and that existing v1 connections start exposing v2 tools. Atlassian's getting started page says actions respect users' existing access controls, recommends least privilege, review of high-impact changes and audit log monitoring, and notes that Teamwork Graph calls such as searchAtlassian consume Rovo credits. Admin settings are documented on a separate Atlassian page.
Cloudflare MCP servers
Cloudflare publishes a recommended Cloudflare API MCP Server at mcp.cloudflare.com/mcp for broad access through code execution, plus product servers for documentation, containers, Browser Run, Digital Experience Monitoring, Cloudflare One CASB and others. Each exposes a stateless Streamable HTTP handler at /mcp; the retired /sse URLs return 410 Gone naming the new URL. The README lists a set of deprecated product servers (including AI Gateway, Audit Logs, DNS Analytics, GraphQL, Logpush, Observability, Radar and Workers servers) and asks users to migrate to the API server.
Context7
Context7, from Upstash, supplies up-to-date, version-specific library documentation and code examples to coding agents. It works in two modes: CLI plus skills, or an MCP server. Setup with npx ctx7 setup authenticates with OAuth and generates an API key; manual configuration uses the server URL with the key in an Authorization: Bearer header. The README recommends a free API key for higher rate limits.
GitHub MCP Server
GitHub hosts a remote server and publishes a local version as a binary and a public container image. Tools are grouped into toolsets; with none specified, the default set is context, repos, issues, pull_requests and users. The GITHUB_TOOLSETS variable or flag narrows or extends this, and --read-only skips write tools even if they are requested. GitHub Enterprise Cloud can use the remote server; GitHub Enterprise Server uses the local server. An insiders mode gives early access to experimental tools.
Notion MCP
Notion's hosted server lets an assistant search a workspace, create and edit pages and manage databases. OAuth is the default; workspace owners on Business and Enterprise plans can create MCP tokens, and Enterprise workspaces with SAML SSO can use enterprise-managed authorisation. Only workspace members can connect, and guests are refused. Notion directs users to the remote server and describes the open source local server, which uses an integration token, as no longer actively maintained; that server's version 2.0 replaced database tools with data source tools.
Playwright MCP
Microsoft's Playwright MCP gives agents browser automation through Playwright's accessibility tree rather than screenshots, and needs Node.js 18 or newer. By default it uses a persistent browser profile; --isolated keeps the profile in memory, --user-data-dir sets its location, --secrets loads values from a dotenv file and --caps enables extra capabilities (vision, pdf, devtools). The README states that --allowed-origins does not serve as a security boundary. It also suggests that coding agents may prefer the Playwright CLI with skills for token efficiency.
Sentry MCP
Sentry's hosted server acts as middleware to the Sentry API and is designed for human-in-the-loop coding agents. Connections can be scoped by URL to an organisation (/mcp/{organizationSlug}) or a project (/mcp/{organizationSlug}/{projectSlug}), and Sentry recommends project scoping where possible. The stdio server in the Sentry CLI uses the CLI login and supports self-hosted Sentry. Its AI-powered search tools need an LLM provider key, and the older npx @sentry/mcp-server launch path is marked legacy.
Stripe MCP
Stripe's hosted server offers API tools (stripe_api_search, stripe_api_details, stripe_api_read, stripe_api_write), account, analytics and documentation tools. OAuth suits interactive clients; agent API keys suit autonomous ones. Stripe states that from December 2026 it will stop accepting full-access secret keys and restricted keys without the Agent tag. Some actions, such as creating a refund, need human confirmation in the Dashboard, and administrators can enable or disable MCP access and revoke OAuth sessions per environment.
Remote or local: decision guidance
| Factor | Remote server | Local server |
|---|---|---|
| Updates | Operated and updated by the provider | You pin and update the package or image |
| Credentials | OAuth tokens held by the client, revocable at the provider | Tokens or keys in local configuration or environment |
| Self-hosted products | Usually cloud editions only (GitHub Enterprise Server and self-hosted Sentry use local) | Can point at self-hosted instances |
| Network | Needs egress to the provider's endpoint; managed networks may need it allowed | Runs on the device; reaches the provider's API directly |
| Visibility for the organisation | Provider-side session lists and admin controls where offered | Requires endpoint inventory of developer machines |
Small teams usually start with remote servers and OAuth because there is nothing to patch. Larger organisations often add a gateway or private subregistry so that only approved servers can be reached; see MCP Gateway vs Direct MCP Connections. The credential trade-off is covered in API Keys vs OAuth for MCP Servers.
How to bring servers into an organisation
- Inventory what is already connected. Read MCP client configuration files on developer machines and list each server, its source and its credential. The MCP Server Inventory Template gives a record format and How to audit MCP servers on developer laptops describes the search.
- Confirm provenance. For each server, confirm it is the publisher's own endpoint or repository, or a reference server, and check the Registry namespace where it is listed.
- Assess before approving. Review tools, scopes and data flows using the MCP Server Trust Assessment.
- Start narrow. Use the narrowest option each provider documents: GitHub toolsets and read-only mode, Sentry project scoping, Stripe agent keys with only needed permissions, Playwright isolated profiles.
- Pilot with a small group. Turn on client tool approval, watch the provider's session and audit views, then widen access.
- Keep up with changes. Track endpoint moves, tool renames and deprecations from the provider's changelog or README, and remove archived reference servers.
Measuring adoption and control
- Approved share: percentage of connected servers that are on the approved list.
- Provenance: number of servers whose publisher cannot be confirmed as the product's own or the MCP project.
- Credential type: share of connections using OAuth or scoped agent tokens rather than broad personal tokens.
- Stale configurations: count of archived reference servers, retired
/sseURLs or v1 endpoints still configured. - Session hygiene: OAuth sessions revoked when people leave or devices are retired.
Common failure patterns
- Installing a look-alike package: the September 2025 postmark-mcp malicious MCP server was an unofficial npm package that copied emails to an attacker; the vendor had never published it.
- Running reference servers as production services: the project describes them as educational examples.
- Keeping archived servers: configurations that still launch the archived GitHub or Sentry reference servers run code the project no longer maintains, instead of the publishers' current servers.
- Default-wide tool exposure: enabling every toolset or all capabilities when a few are needed.
- Treating origin allowlists as a boundary: Playwright MCP's own documentation says
--allowed-originsis not a security boundary. - Missing endpoint changes: clients left on retired URLs or old key types fail, and teams sometimes fix them by switching to broader credentials.
For hardening detail, see the MCP Security Guide, the Local MCP Server Security Guide and the Remote MCP Server Hardening Guide.
Practitioner checklist
- Find servers through the MCP Registry or the publisher's own documentation, not search results for a package name.
- Confirm each server is a reference server or the product owner's own, and record its namespace and source.
- Prefer OAuth for interactive use and scoped agent tokens for automation; avoid broad personal tokens.
- Apply each provider's narrowing options: toolsets, read-only mode, organisation or project scoping, isolated profiles.
- Turn on tool approval in the MCP client and use provider-side confirmation where offered.
- Replace archived reference servers, retired
/sseURLs and v1 endpoints. - Pin versions of local servers and review release notes before upgrading.
- Review and revoke OAuth sessions at each provider on a schedule and at leaver events.
Standards and references
- Model Context Protocol: servers repository (reference servers)
- Model Context Protocol blog: Introducing the MCP Registry
- Model Context Protocol: registry repository and publishing authentication docs
- Atlassian: Getting started with the Atlassian Rovo MCP Server
- Cloudflare: MCP servers repository
- Upstash: Context7 repository
- GitHub: GitHub MCP Server repository
- Notion: Get started with Notion MCP
- Microsoft: Playwright MCP repository
- Sentry: Sentry MCP
- Stripe: Model Context Protocol documentation
Related NHI Mgmt Group resources: What Is MCP? The Model Context Protocol Explained (2026) · MCP Client Support Tracker · MCP Registry and Server Publishing Security Guide · MCP Server Security Checklist · AI Agent Frameworks and Platforms Hub (2026)