Which MCP Servers Connect AI to Developer Tools?

Developer-tool MCP servers expose code platforms, issue trackers, and error monitoring systems to AI assistants through the Model Context Protocol (MCP). Anthropic created MCP and open-sourced it in November 2024. These servers let Claude, ChatGPT, Cursor, and VS Code interact with GitHub repositories, GitLab merge requests, Linear tickets, and Sentry error logs without custom integrations.

Six vendor-maintained developer-tool servers exist as of mid-2026. Each vendor builds and maintains its own server. The archived Anthropic reference implementations for GitHub and GitLab are no longer updated. Install the vendor versions.

This page covers what developer-tool MCP servers expose, which servers are available, how to choose between them, and when a CLI is the better option.

What Do Developer-Tool MCP Servers Expose?

Developer-tool MCP servers expose repository operations, issue tracking, pull request management, CI/CD pipelines, and error monitoring to any MCP-compatible AI application.

The category covers four types of developer infrastructure. Code platforms (GitHub, GitLab, Bitbucket) expose repositories, branches, commits, file contents, diffs, and code search. Issue trackers (Linear, Jira) expose tickets, projects, cycles, labels, and workflow states. Error monitoring (Sentry) exposes production errors, stack traces, frequency trends, and affected user counts. CI/CD systems are exposed as a subset of the code platform servers. GitLab’s MCP server provides pipeline control. GitHub’s exposes workflow status.

All developer-tool servers expose MCP tools (actions the model can execute). Most do not expose MCP resources (read-only data streams) or MCP prompts (reusable templates). The interaction model is tool-based: the model calls a tool, the server runs it against the platform’s API, and the result returns to the conversation.

The practical value is workflow automation. A model that connects to both a code platform and an issue tracker can read a pull request, identify the issues it addresses, update ticket statuses, and post a summary. That workflow replaces four manual context switches with a single conversation.

Which Developer-Tool MCP Servers Are Available?

Six vendor-maintained developer-tool MCP servers cover the primary platforms: GitHub, GitLab, Linear, Sentry, Jira (Atlassian), and LaunchDarkly.

Each capsule below describes what the server exposes, who maintains it, and the authentication model. Vendor-maintained servers receive regular updates.

GitHub

The GitHub MCP server exposes repositories, issues, pull requests, branches, commits, and code search across public and private repositories. GitHub maintains this server. It registers 93 tool definitions organized into modular toolsets. Two versions exist: a local Docker image authenticated with a personal access token, and a remote hosted endpoint at api.githubcopilot.com/mcp/ authenticated through OAuth. The remote version requires a GitHub Copilot seat. The local version requires Docker.

Token cost: 17,600 to 55,000 tokens for tool definitions, depending on enabled toolsets. Disable unused toolsets to reduce overhead.

GitHub MCP server setup guide, token scopes, and costs

GitLab

The GitLab MCP server exposes projects, merge requests, issues, CI/CD pipelines, wiki pages, releases, tags, and milestones. GitLab maintains this server. It supports three transport types: stdio for local use, SSE for legacy clients, and streamable HTTP for remote deployments. Authentication options include personal access tokens, local OAuth 2.0, and remote MCP OAuth with dynamic client registration. The server works with both GitLab.com and self-managed GitLab instances.

The archived Anthropic reference GitLab server at modelcontextprotocol/servers is no longer updated. Install the GitLab-maintained version: gitlab/mcp-server on npm or gitlab/mcp-server:latest on Docker.

Linear

The Linear MCP server exposes issues, projects, cycles, comments, labels, and roadmap data through a centrally hosted remote endpoint. Linear maintains this server. It uses streamable HTTP transport with OAuth 2.1 and dynamic client registration. No local installation needed. No Docker. No personal access token in a config file. Connect by adding the remote URL to your client configuration. Claude Desktop, Cursor, and VS Code support it natively.

Linear’s MCP server is considered the reference standard for B2B SaaS MCP implementations. The tool surface mirrors Linear’s GraphQL API but is purpose-shaped for AI consumption: descriptive tool names, structured outputs, and low-cardinality enums.

Sentry

The Sentry MCP server exposes error tracking data: recent errors, stack traces, frequency trends, affected user counts, and release associations. Sentry maintains this server. A remote endpoint launched in early 2026. The practical query: “are there any new errors since yesterday’s deploy?” replaces a manual dashboard check every morning.

Sentry also launched MCP Server Monitoring in August 2025: a separate product that monitors the health of MCP servers themselves. The monitoring tool and the MCP server are different products. The MCP server exposes Sentry data to AI assistants. The monitoring tool tracks whether MCP servers are running correctly.

Jira and Confluence (Atlassian)

Atlassian’s MCP servers expose Jira issues, Confluence pages, and project data through remote HTTP endpoints. Atlassian maintains these servers. SSE transport is deprecated as of June 30, 2026. Streamable HTTP is the supported transport. Authentication uses OAuth 2.1.

Atlassian’s servers connect AI assistants to the same Jira and Confluence data that development and product teams work in daily. The model can search issues, create tickets, read Confluence documentation, and update project status.

LaunchDarkly

The LaunchDarkly MCP server exposes feature flag management: flag listing, targeting rules, flag status, and environment configuration. LaunchDarkly maintains this server. The primary use case is debugging: “is the new checkout flow enabled for enterprise accounts?” answers from live flag data instead of requiring a dashboard login.

How Do You Choose Between Developer-Tool Servers?

Choose based on what your team already uses. Each server connects to one platform. Install the server that matches your existing code platform, issue tracker, or monitoring tool.

A team on GitHub and Linear installs the GitHub and Linear MCP servers. A team on GitLab and Jira installs the GitLab and Atlassian servers. No decision framework is needed when the platform choice is already made.

The real decision is how many to run simultaneously. Each server’s tool definitions consume context window space. The GitHub server alone costs 17,600 to 55,000 tokens. Three developer-tool servers running together consume 30,000 to 80,000 tokens of definitions before the first message. Three to five total servers (across all categories, not just developer tools) is the practical ceiling for most models.

Connect only the servers the current task needs. A code review session needs GitHub. It does not need Linear or Sentry. An error investigation needs Sentry and the code platform. It does not need the issue tracker until the model creates a ticket for the fix.

The evaluation criteria for maintenance recency, authentication model, and transport support apply to developer-tool servers the same way they apply to every other category.

Developer-tool servers expose code platforms, ticket systems, and error logs. The workflows that combine them with servers from other categories (databases, communication, file systems) are where the productivity gains compound.

When Is a CLI Better Than a Developer-Tool MCP Server?

A CLI is better for single operations. An MCP server is better for chained operations where the model needs to reason across multiple steps.

The GitHub CLI (gh) lists issues, checks PR status, and creates branches in one command. Each operation costs zero context window tokens. The equivalent MCP tool call costs 500 to 2,000 tokens for the call and response, on top of the 17,600+ tokens for tool definitions. For a developer running gh pr list once, the CLI is 4 to 32 times cheaper per operation (OnlyCLI benchmark, March 2026).

The MCP server earns its cost when the model chains operations. “Review the open PRs, find the ones that touch the auth module, check for issues related to each, and post a summary.” That workflow requires reading PR diffs, searching issues, correlating them, and composing a summary. A CLI requires the developer to script each step. An MCP server lets the model reason through the chain.

The boundary is not binary. GitHub itself uses both approaches in production, cutting token costs in agentic CI workflows by up to 62% through selective MCP-to-CLI replacement for simple operations while keeping MCP for multi-step chains (InfoQ, May 2026).

Developer-Tool MCP Server Questions

Are Developer-Tool MCP Servers Free?

Yes. Every vendor-maintained developer-tool MCP server listed on this page is free and open-source. The MCP protocol itself is free under the Linux Foundation. No licensing fees. No usage charges. The only costs are the external platform’s own limits: GitHub API rate limits, Linear API quotas, Sentry plan thresholds.

Can You Use Multiple Developer-Tool Servers at Once?

Yes. MCP clients support multiple simultaneous servers. A Claude Desktop configuration can include GitHub, Linear, and Sentry in the same config file. Each server runs as a separate process. The model sees all tools from all connected servers. The constraint is context window capacity: each server’s tool definitions consume tokens. Connect only the servers the current workflow requires.

Do Developer-Tool MCP Servers Work in All Clients?

All six vendor-maintained servers listed on this page work in Claude Desktop, Cursor, and VS Code. Remote servers (Linear, Sentry, Atlassian) use streamable HTTP, which requires client support for that transport. Claude Desktop, Cursor, VS Code, and Windsurf all support streamable HTTP as of mid-2026. Local servers (GitHub Docker version, GitLab stdio version) use stdio, which every MCP client supports.

Set up MCP servers in Claude Desktop · All MCP servers by category