Skip to main content

MCP Servers for Developers: Official Options

Short answer

The vendors behind the tools you already use publish MCP servers for them. GitHub, GitLab, Atlassian (Jira and Confluence), Linear, Sentry, Cloudflare, Vercel, PagerDuty and Microsoft's Azure DevOps all have one, and the MCP project maintains a reference Git server for local repositories. Most vendor servers are remote and sign in with OAuth. GitHub and Linear can run read-only, and GitHub's local server has a --read-only flag. Start with read-only access, and give write access only when you need it.

Tested withGitHub MCP Server README (checked 2026-10-20)GitLab MCP server documentation (checked 2026-10-20)Atlassian MCP Server README (checked 2026-10-20)Linear MCP documentation (checked 2026-10-20)Sentry MCP documentation and README (checked 2026-10-20)Cloudflare MCP servers documentation (checked 2026-10-20)Vercel MCP documentation (checked 2026-10-20)PagerDuty MCP documentation (checked 2026-10-20)Azure DevOps MCP Server README (checked 2026-10-20)MCP reference Git server README (checked 2026-10-20)Checked 20 Oct 2026

By Tony LewisHow we test these guides

What's in this list​

The list covers source control, issue trackers, CI, error monitoring and cloud consoles. Each entry names the publisher, how the server runs, how it signs in, whether a read-only mode is documented and one risk to check. Every entry links to the vendor's documentation or repository, and we checked each one on 2026-10-20.

Some tools a developer might expect aren't here. CircleCI's older community server is deprecated, and we didn't verify a replacement, so it's left out. Datadog and the major cloud providers' consoles weren't verified in the same way, so they're left out too. Check the publisher's own docs for those.

Source control​

Git (reference server)​

The Git server in the modelcontextprotocol/servers repository is a reference implementation from the MCP project. It runs locally, and you start it with uvx mcp-server-git --repository <path>, where <path> is any folder inside a Git repository. It reads status, diffs and logs, and it can also run git_commit, git_reset and git_checkout.

There is no read-only switch. If you only want the model to read history, don't give it a repository you need to keep safe, or run it on a clone.

Risk: the server runs with your user's permissions, so a commit or reset changes your working copy. Read the tool list before you use it on a repository with uncommitted work.

Source: github.com/modelcontextprotocol/servers, with the Git server's README at src/git

GitHub​

GitHub's MCP server is available as a remote server at https://api.githubcopilot.com/mcp/. Your AI app connects with OAuth, or with a GitHub personal access token set as GITHUB_PERSONAL_ACCESS_TOKEN if the app doesn't support OAuth for remote servers. The README says a token takes precedence over OAuth when both are set.

The same project also runs locally from the ghcr.io/github/github-mcp-server Docker image. Its README covers OAuth login in Docker through a fixed callback port, and it says the login token stays in memory only. Read-only mode is documented: start the server with --read-only, and write tools are removed from the tool list.

Risk: a personal access token carries the permissions you grant it, so give it only the repositories and scopes you need. With --read-only off, the server can also change repositories, issues and pull requests, as far as the account allows.

Source: github.com/github/github-mcp-server

GitLab​

GitLab runs an MCP server on GitLab instances, including GitLab.com, at https://<your-gitlab-host>/api/v4/mcp. It has been generally available since GitLab 19.5, and an administrator has to allow access to it for the top-level group on GitLab.com or for the instance on self-managed GitLab. Clients sign in with OAuth 2.0 dynamic client registration, so the AI app registers itself with your GitLab instance on first connect. You can also use a fine-grained personal access token that has the MCP tool permission, but legacy, project and group access tokens are rejected.

The server also has toolsets you can switch on or off with the X-Gitlab-Enabled-Mcp-Server-Toolsets header. The documentation doesn't describe a single read-only switch.

Risk: the server acts with the permissions of the account that signs in. Sign in with an account that only has access to the projects you want the assistant to see.

Source: GitLab MCP server documentation, with the source file at doc/user/model_context_protocol/mcp_server.md

Issue trackers​

Atlassian (Jira and Confluence)​

Atlassian's official MCP server is a cloud-hosted server at https://mcp.atlassian.com/v2/mcp. The README recommends v2 for new setups. Sign in with OAuth 2.1, or with an API token for headless use. Access follows the permission groups you grant for each product, and the README links to a supported-tools page that lists the scope each tool needs.

The README doesn't describe a single read-only switch for the whole server. Use the permission groups to grant only the read access you need.

Risk: a token or OAuth grant that includes write permission lets the assistant create and change issues and pages. Review the permission groups on the OAuth screen before you accept them.

Source: github.com/atlassian/atlassian-mcp-server

Linear​

Linear runs a remote server at https://mcp.linear.app/mcp. You sign in with OAuth. Linear's documentation also describes a read-only endpoint, https://mcp.linear.app/mcp/readonly, which only ever exposes read tools. A client that requests only the read OAuth scope on the standard endpoint is also limited to read access.

Risk: the standard endpoint with the write scope lets the assistant create and update issues, projects and comments in every team you can access. Use the read-only endpoint for search and triage.

Source: linear.app/docs/mcp

Error monitoring​

Sentry​

Sentry runs a remote server at https://mcp.sentry.dev/mcp. Every connection uses OAuth, and the first connection opens a sign-in to your Sentry account. You can scope the URL to one organization, https://mcp.sentry.dev/mcp/{organizationSlug}, or to one project as well. The Sentry project's README also describes a locally run server, npx @sentry/mcp-server@latest --access-token=<token>, for a user token.

Sentry's README describes skills, which are groups of tools. You can limit a server to some of them with --skills, or remove one with --disable-skills. The README doesn't describe a separate read-only switch.

Risk: a local server with a user token can act with that token's permissions, so create a token with only the scopes the assistant needs. Keep the skills you don't use switched off.

Sources: docs.sentry.io/product/sentry-mcp, github.com/getsentry/sentry-mcp

PagerDuty​

PagerDuty runs a remote server at https://mcp.pagerduty.com/mcp. It accepts API tokens and OAuth, and the documentation links PagerDuty's OAuth metadata for clients that support it. The documentation for the remote server doesn't describe a read-only mode.

PagerDuty's earlier local server, PagerDuty/pagerduty-mcp-server, is now deprecated and read-only. Its README says it exposes only read-only tools by default, and that you add --enable-write-tools to allow tools that change data. That's a local server with its own setup, so check the remote server's documentation for what it can change.

Risk: the remote server can see incidents and services, and it can change them if the token or OAuth grant allows it. Treat incident data as sensitive, and check which permissions your token or grant includes before you connect.

Sources: developer.pagerduty.com/docs/mcp, github.com/PagerDuty/pagerduty-mcp-server

CI and build pipelines​

We couldn't verify a current CI server from most CI vendors, so this section is short. For Azure Pipelines, use the Azure DevOps server below. For other CI systems, check the vendor's own docs.

Azure DevOps​

Microsoft publishes the Azure DevOps MCP Server, which runs locally with npx -y @azure-devops/mcp <organization>. Sign in with a Microsoft account that can access the organization. The README also documents --authentication azcli for sign-in through the Azure CLI.

The server loads every tool group unless you choose domains with -d, which keeps the tool list short. The README lists nine domains, including core, work-items, repositories, pipelines and wiki. It doesn't describe a single read-only switch, so choose only the domains you need.

Risk: the sign-in gives the server the same access to projects, repositories and pipelines as your account. Use a separate account with narrow access if the assistant doesn't need everything you can see.

Source: github.com/microsoft/azure-devops-mcp

Cloud consoles​

Cloudflare​

Cloudflare runs managed remote servers. The Cloudflare API MCP server at https://mcp.cloudflare.com/mcp gives access to the Cloudflare API, and Cloudflare also runs product servers such as https://docs.mcp.cloudflare.com/mcp. You connect with OAuth, and the Cloudflare authorization screen lets you choose the permissions to grant.

Risk: the API server covers your whole account, so the permissions you grant on the OAuth screen decide what the assistant can change. Grant read permissions first, and add write permissions only for a task that needs them.

Source: Cloudflare MCP servers documentation

Vercel​

Vercel runs a remote server at https://mcp.vercel.com. The Vercel documentation describes it as Vercel's official MCP server, and clients sign in with OAuth. The documentation doesn't describe a read-only switch. Its own examples show read-only use, such as reading deployment data to diagnose a failed deployment, but that comes from how you set up the agent, not a server option.

Risk: the server gives the assistant access to your Vercel projects, so review the permissions and data access level of each agent and tool before you approve the connection.

Source: vercel.com/docs/mcp/vercel-mcp

Search the MCPBundles directory​

To search for a developer tool by name, open the MCPBundles directory and type the name in the search box, or go straight to a search such as www.mcpbundles.com/?q=github. Each server has a listing page at https://www.mcpbundles.com/skills/<slug> with the setup steps for your AI app.

Listings are third-party, so check the publisher before you connect one. Discovering MCP servers covers the filters, and evaluate an MCP server has the checklist to run first.

FAQ​

Which GitHub MCP server should I use?

GitHub publishes a remote server at api.githubcopilot.com/mcp/ that signs in with OAuth, and a local version that runs from the ghcr.io/github/github-mcp-server Docker image. Add --read-only to the local version if you only need to read repositories and issues.

Can an MCP server change my issues or pull requests?

Yes, if the server exposes write tools and your sign-in allows writes. Use a read-only mode where the server has one, and check the permission groups or scopes on the sign-in screen.

Is the GitLab MCP server ready for production use?

Yes. GitLab moved the server from beta to generally available in GitLab 19.5. It runs on GitLab.com, self-managed and Dedicated instances once an administrator allows access, and it uses OAuth or a fine-grained personal access token.

Why is there no CircleCI or Datadog server on this page?

We didn't verify a current server from either vendor in the sources we checked, so we left them out. Check the vendor's own documentation before you search for one.