MCP Servers for Coding Agents: 10 Worth Installing
The 10 MCP servers worth installing for coding agents in 2026: GitHub, Context7, Playwright, Chrome DevTools, Sentry, Supabase, Linear, Jira, Figma, Memory.
Assumer Team ·
Short answer: for most developers, the MCP servers worth installing are GitHub (issues and pull requests), Context7 (up-to-date library docs), Playwright and Chrome DevTools (browser testing and debugging), Sentry (production errors), Supabase (database and backend), Linear or Atlassian (tickets), Figma (design context) and Memory (persistent notes). Install only the ones that match your stack, prefer the vendor’s official server, and start with read-only access.
What MCP is, in one paragraph
The Model Context Protocol is an open standard for connecting AI applications to external tools and data. A coding agent such as Assumer, OpenCode, Claude Code, Cursor or Gemini CLI acts as the client; an MCP server exposes tools, like “create issue” or “run SQL query”, that the agent can call. Servers run either locally (started by the agent, usually through npx or Docker) or remotely (hosted by the vendor, reached over HTTP with OAuth or a token).
Every server below was checked against its official repository or documentation in October 2026.
The 10 MCP servers at a glance
| # | Server | Use it for | Type | Official source |
|---|---|---|---|---|
| 1 | GitHub | Repos, issues, PRs, Actions | Remote or local (Docker) | github/github-mcp-server |
| 2 | Context7 | Version-specific library docs | Remote or local | upstash/context7 |
| 3 | Playwright | Browser automation and testing | Local (npx) | microsoft/playwright-mcp |
| 4 | Chrome DevTools | Debugging, network, performance | Local (npx) | ChromeDevTools/chrome-devtools-mcp |
| 5 | Sentry | Production errors and issues | Remote or local | getsentry/sentry-mcp |
| 6 | Supabase | Database, schema, project config | Remote | supabase-community/supabase-mcp |
| 7 | Linear | Issues, projects, cycles | Remote | Linear MCP docs |
| 8 | Atlassian | Jira and Confluence | Remote | atlassian/atlassian-mcp-server |
| 9 | Figma | Design context for UI code | Remote or desktop | Figma MCP docs |
| 10 | Memory | Persistent knowledge graph | Local (npx) | modelcontextprotocol/servers |
The servers in detail
1. GitHub MCP server
GitHub’s official server lets the agent read and act on repositories, issues, pull requests, Actions workflows, code scanning alerts and more. It is available as a hosted remote server at https://api.githubcopilot.com/mcp/ or locally through the ghcr.io/github/github-mcp-server Docker image. You can limit it to the toolsets you need, such as repos, issues and pull_requests.
Good for: “Read issue #412, implement it, and open a pull request that references it.”
2. Context7
Models are trained on a snapshot of the web, so they often suggest outdated APIs. Context7 pulls version-specific documentation and code examples into the prompt. It is available remotely at https://mcp.context7.com/mcp or locally as @upstash/context7-mcp; a free API key raises rate limits.
Good for: working with fast-moving frameworks where last year’s API is wrong.
3. Playwright MCP
Microsoft’s Playwright MCP lets the agent drive a real browser. It works from the page’s accessibility tree rather than pixels, so it does not need a vision model.
npx @playwright/mcp@latestGood for: “Open the signup page, try an invalid email, and confirm the error message appears.”
4. Chrome DevTools MCP
The Chrome DevTools MCP server exposes DevTools to the agent: performance traces, network requests, console messages with source-mapped stack traces, and screenshots.
npx -y chrome-devtools-mcp@latestGood for: “Why is the dashboard slow to load?” with an actual trace to back up the answer.
5. Sentry MCP
Sentry’s server connects the agent to your error tracking. Use the hosted endpoint at https://mcp.sentry.dev/mcp, or run @sentry/mcp-server locally with an access token, which also works with self-hosted Sentry.
Good for: “Look at the top unresolved error in the checkout project and propose a fix.”
6. Supabase MCP
The Supabase MCP server is hosted at https://mcp.supabase.com/mcp. Two URL options are worth using from day one: read_only=true to block mutating tools, and project_ref=<ref> to scope the server to one project. Supabase recommends reading its security guidance before connecting an LLM to a project.
Good for: inspecting schemas and writing migrations against a development project.
7. Linear MCP
Linear runs an official remote server at https://mcp.linear.app/mcp with OAuth sign-in, as described in its docs. A read-only endpoint (/mcp/readonly) is also available.
Good for: “Summarise the bugs assigned to me this cycle and draft an implementation plan with sub-issues.”
8. Atlassian MCP (Jira and Confluence)
The Atlassian remote server covers Jira, Confluence and Compass. Its recommended endpoint is https://mcp.atlassian.com/v2/mcp, with OAuth 2.1 or admin-enabled API tokens. Delete and manage tools for Jira are disabled by default.
Good for: teams whose specs and tickets live in Confluence and Jira.
9. Figma MCP
Figma’s server gives coding agents design context: components, variables, layout data and code generation from a selected frame. The remote server is hosted at https://mcp.figma.com/mcp; a desktop server for specific organisation use cases runs at http://127.0.0.1:3845/mcp when enabled in Dev Mode. Figma recommends the remote version for most users.
Good for: turning a design frame into components that match your existing design system.
10. Memory
The Memory server from the official MCP reference servers stores entities, relations and observations in a local knowledge graph, so facts can persist across sessions.
npx -y @modelcontextprotocol/server-memoryGood for: remembering decisions and project facts that do not belong in a rules file.
What you probably do not need
Some popular servers duplicate tools your agent already has. Assumer, for example, already includes read, write, edit, grep, glob, list, bash and webfetch tools, so a generic filesystem or fetch server adds little. Note also that several older reference servers, including GitHub, PostgreSQL, Puppeteer, Sentry and Slack, have been moved to the archived reference list in the modelcontextprotocol/servers repository. Use the vendor-maintained servers listed above instead.
Adding MCP servers to Assumer
npm i -g @4born-solutions/assumer
assumer login
assumer mcp addassumer mcp add walks you through adding a local (stdio) server, such as npx @playwright/mcp@latest, or a remote (HTTP) server, such as https://mcp.linear.app/mcp, including OAuth where the server supports it. Inside a session, /mcps lets you manage connected servers.
Then set permissions. A sensible default:
| Kind of tool | Permission |
|---|---|
| Read-only (search docs, list issues, read errors) | allow |
| Writes (create issue, comment, run SQL) | ask |
| Destructive (delete, drop, force-push) | deny |
Best practices for MCP in coding agents
Keep the list short
Every server adds tool descriptions to the agent’s context. Enable what you use; turn off the rest per project.
Prefer official servers
Use the vendor’s own server or the official reference servers. Review community servers before running them, because a local MCP server runs code on your machine.
Use least privilege
Use read-only modes, project scoping and fine-grained tokens. Never connect production databases with write access to an agent.
Document servers in your rules file
Add a line to AGENTS.md or Assumer.md, such as “Use Context7 for library docs before writing integration code”, so the agent knows when to reach for each tool.
FAQ
What is an MCP server for a coding agent?
An MCP server exposes tools and data to an AI agent through the Model Context Protocol, an open standard. For a coding agent that might mean reading GitHub issues, querying a database, driving a browser or fetching current library documentation.
How many MCP servers should I install?
Only the ones you use regularly. Each server adds tool descriptions to the agent’s context, so a handful of relevant servers usually works better than dozens. Enable project-specific servers per project.
Are MCP servers safe?
They are as safe as the access you grant. Prefer official servers from the vendor, use read-only modes and scoped tokens where offered, keep write actions on ask, and never give a server production credentials it does not need.
What is the difference between local and remote MCP servers?
Local servers run on your machine, usually started with npx or Docker over stdio. Remote servers are hosted by the vendor and reached over HTTP, often with OAuth sign-in, so there is nothing to install or update locally.
How do I add an MCP server to Assumer?
Run assumer mcp add in your terminal and follow the prompts for a local command or a remote URL. You can then manage servers with the /mcps command and control their tools with allow, ask or deny permissions.