ScreenshotNeo

BlogComparisons

The Best Free MCP Servers for Cursor in 2026

Compare the best free-to-start MCP servers for Cursor—Context7, GitHub and Atlassian—with setup steps, permissions and limits.

By the ScreenshotNeo team1 October 20269 min read

Short answer: For most Cursor users, start with Context7 for current library documentation, GitHub’s official MCP server for repository work, or Atlassian’s official MCP server for Jira and Confluence. These are free to start in the ways documented by their vendors, but an MCP server being free or offering a free key does not prove that the connected provider is unlimited or free at every usage level.

Cursor connects MCP servers through its Marketplace or a custom mcp.json file. The right choice depends on the system your agent needs to read or change, how you want to authenticate, and the limits of the connected service. This guide gives current setup examples, permission advice, troubleshooting steps and a practical comparison.

1. What MCP adds to Cursor

Model Context Protocol (MCP) lets Cursor’s agent call external tools and access external data. A server can expose documentation search, repository operations, issue data or other actions that are unavailable in the editor by itself. Cursor documents three transport styles: local stdio, remote or local SSE, and Streamable HTTP. You can install supported servers from the Cursor Marketplace or define your own server in mcp.json. Cursor’s MCP documentation explains the current configuration and security model.

Connected servers may access external systems and perform actions for you. Verify the server source, review every permission, use least-privilege API keys, and audit code when an integration is critical.

2. Best free-to-start MCP servers at a glance

Server Best for Connection and authentication What “free” means here
Context7 Current library and framework documentation Hosted MCP endpoint or CLI; optional Context7 API key Vendor documents a free key for higher rate limits; check current service limits
GitHub official MCP server Repositories and GitHub workflows Hosted Streamable HTTP with a GitHub personal access token, or local Docker with OAuth Official server and setup are available; GitHub account, token scope and plan rules still apply
Atlassian official MCP server Jira and Confluence work Recommended v2 endpoint; OAuth 2.1 or API token Requires an Atlassian Cloud site and product access; applicable plan terms apply

Recommendation: choose by workflow rather than by the word “free.” Context7 improves answers about a package or framework version. GitHub is the natural choice for repository work. Atlassian fits teams already using Jira or Confluence.

3. Context7: current documentation for libraries and frameworks

Context7 supplies current code examples and documentation to coding agents. Its repository documents two approaches: a CLI plus skills, or MCP for native tool calls. The CLI setup supports Cursor and requires Node.js 18 or newer. A free API key is recommended for higher rate limits; that is not a promise of unlimited requests.

Install with the documented setup command

npx ctx7 setup

Follow the prompts and select Cursor as the target. When asking Cursor a question, include the library identifier or version when you know it. That gives the server a narrower target and reduces ambiguous documentation matches. See the Context7 repository for the current CLI and MCP instructions.

When Context7 is the right choice

  • You need examples for a particular framework release.
  • Your agent is guessing from stale training data.
  • You want documentation retrieval as a native tool call instead of copying pages into chat.

4. GitHub’s official MCP server: repository workflows

GitHub publishes an official MCP server and a Cursor-specific setup guide. The hosted server is https://api.githubcopilot.com/mcp/. The hosted setup requires Cursor 0.48.0 or later for Streamable HTTP and a GitHub personal access token. The project also documents a local Docker route that can use OAuth.

Hosted Streamable HTTP configuration

In Cursor, add a custom server through the MCP settings or your project/user mcp.json. The exact UI and schema can change, so use GitHub’s current guide when copying the final fields. A representative configuration is:

{
  "mcpServers": {
    "github": {
      "url": "https://api.githubcopilot.com/mcp/",
      "headers": {
        "Authorization": "Bearer YOUR_GITHUB_TOKEN"
      }
    }
  }
}

Use a token scoped only to the repositories and actions the agent needs. Never commit this file with a real secret. Cursor’s environment-variable support can keep credentials outside the project; follow the current Cursor and GitHub instructions for the supported syntax.

Local Docker option

GitHub’s official repository documents a local Docker setup with OAuth. Use that route when you need the server process and authentication flow to stay on your machine. Do not follow older tutorials that install @modelcontextprotocol/server-github; GitHub says that npm package is no longer supported as of April 2025. Start from the official GitHub MCP repository and its Cursor setup guidance.

Good GitHub tasks for an agent

  • Find files, issues, pull requests or repository documentation.
  • Summarize changes across a branch or pull request.
  • Open or update issues when the token and approval policy allow it.

Review a proposed write action before allowing it. A read-only token is safer for investigation and code review.

5. Atlassian MCP server: Jira and Confluence

Atlassian’s official MCP server supports Cursor and connects to Atlassian Cloud products, including Jira and Confluence. The repository recommends the v2 endpoint:

https://mcp.atlassian.com/v2/mcp

Authentication can use OAuth 2.1 or an API token, depending on the setup. Tool access follows the permissions of the Atlassian user. You need an Atlassian Cloud site and access to the relevant product. Check the official Atlassian repository for the current Cursor configuration and authentication flow.

Choose Atlassian when

  • Your work already lives in Jira projects or Confluence spaces.
  • The agent must search specifications, tickets or runbooks alongside code.
  • Your administrator can grant narrowly scoped access to the required sites and projects.

Existing Atlassian permissions still apply. MCP does not bypass project, space or account access controls.

6. How to add an MCP server to Cursor

  1. Choose the workflow. Pick documentation, GitHub or Atlassian based on the data and actions your task needs.
  2. Use the vendor’s official source. Confirm the repository owner, endpoint and current authentication instructions.
  3. Install from Marketplace or configure mcp.json. Marketplace installation is convenient; JSON is useful for a custom endpoint or a project-specific setup.
  4. Authenticate with least privilege. Prefer read-only scopes while evaluating a server. Keep tokens in environment variables or the provider’s OAuth flow.
  5. Restart or reload Cursor. The MCP panel should show the server and its tools.
  6. Run a harmless read operation. Ask for a documentation lookup, repository listing or Jira search before permitting writes.
  7. Audit the result. Confirm that the tool used the expected account, repository, site and scope.

7. Comparing cost, limits and operational fit

An MCP server is a connection layer. It does not define the price or quota of the service it calls. Before adopting one, check:

  • Provider account: Is a GitHub, Atlassian or Context7 account required?
  • Usage limits: Does the provider cap requests, tokens, projects or rate?
  • Authentication: Is OAuth available, or must you manage a long-lived token?
  • Transport: Does the server run locally, or does Cursor call a remote endpoint?
  • Write permissions: Can the agent create, edit or delete data?
  • Team policy: Are external connections approved for your repository and workspace?

Measure the tasks you actually run for a week before assuming a free tier is sufficient. Documentation lookups, repository searches and issue updates have different request patterns.

8. Performance and reliability practices

  • Ask for a specific library, version, repository, project or Confluence space to reduce broad searches.
  • Keep prompts focused so the agent makes fewer tool calls.
  • Use local Docker when a remote connection is slow or your policy requires local execution.
  • Expect authentication expiry and rotate tokens through the provider’s supported flow.
  • Keep a manual fallback: the provider’s web UI or documentation remains useful when an MCP endpoint is unavailable.
  • For write operations, require a preview or diff before approval.

9. Troubleshooting common errors

Symptom Likely cause Fix
Server does not appear in Cursor Invalid JSON, wrong file location or an outdated Cursor build Validate mcp.json, reload Cursor and check the MCP panel. For GitHub hosted Streamable HTTP, use Cursor 0.48.0 or newer.
401 or 403 from GitHub Missing, expired or under-scoped personal access token Create or refresh a token with only the required scopes and confirm the Authorization header is passed as documented.
GitHub tutorial references an npm package Old instructions Do not install @modelcontextprotocol/server-github; use the current official GitHub project.
Context7 returns weak or unrelated examples Library or version was not specified Include the library ID and exact version in the prompt, then retry.
Context7 rate limit Free request allowance or anonymous limit reached Use the documented free API key option and review the provider’s current limits.
Atlassian tools show no projects or pages The authenticated user lacks product, site, project or space permission Sign in with the intended Atlassian account and ask an administrator to grant the minimum required access.
OAuth callback fails Redirect URL or local Docker configuration does not match the provider setup Follow the provider’s current OAuth and redirect configuration exactly; remove stale callback entries.
Tool calls are unexpectedly slow Remote latency, broad search or repeated calls Narrow the prompt, reduce result scope and consider a local transport where supported.

10. Security checklist

  • Verify the source repository and endpoint before installing.
  • Read the tool list and understand which operations can mutate data.
  • Use separate, narrowly scoped credentials for agents.
  • Never paste production secrets into prompts or commit them to mcp.json.
  • Review OAuth grants and revoke unused connections.
  • Audit server code for critical or sensitive integrations.
  • Log important write actions and require human approval where appropriate.

11. Or skip the browser setup

If your Cursor workflow needs screenshots of documentation, issues or rendered pages, ScreenshotNeo provides an MCP server with take_screenshot, get_page_info and capture_pdf tools for Claude, Cursor and other MCP clients. You can also call its HTTP API directly.

The API accepts one GET request and returns PNG, JPEG, WebP or PDF. Before capture it accepts cookie or consent banners and removes more than 60 known consent platforms, newsletter popups and chat widgets; each step can be disabled. Bot checks, blank pages, timeouts, failed loads and cache hits are not billed, and the response identifies the result with X-Page-Verdict and X-Billed headers.

cURL

curl -G "https://api.screenshotneo.com/v1/shot" -d access_key=YOUR_API_KEY --data-urlencode url=https://stripe.com -o shot.webp

Python

import requests

r = requests.get(
    "https://api.screenshotneo.com/v1/shot",
    params={"access_key": "YOUR_API_KEY", "url": "https://stripe.com"},
    timeout=90,
)
r.raise_for_status()
open("shot.webp", "wb").write(r.content)

Node.js

const q = new URLSearchParams({ access_key: 'YOUR_API_KEY', url: 'https://stripe.com' });
const res = await fetch(`https://api.screenshotneo.com/v1/shot?${q}`);
if (!res.ok) throw new Error(`HTTP ${res.status}`);
const body = Buffer.from(await res.arrayBuffer());
require('fs').writeFileSync('shot.webp', body);

See the ScreenshotNeo API documentation for the 63 available options, including full-page and element capture, device presets, dark mode, retina scale, custom CSS and JavaScript, waits, blocked resources, headers and cookies, geolocation, PDF settings, caching, signed links, async webhooks, bulk capture and usage reporting. Plans include 1,000 free shots per month with no card; paid plans start at $5 for 3,000 shots.

Create a free ScreenshotNeo account and try the MCP server or API with 1,000 screenshots each month at no charge.

12. FAQ

Is an MCP server itself an API plan?

No. MCP exposes tools to Cursor; the connected provider still controls authentication, pricing and quotas.

Which server should a solo developer install first?

Use Context7 if documentation quality is the main problem. Add GitHub when repository context is central. Add Atlassian only when your work depends on Jira or Confluence.

Can I install all three?

Yes, provided you understand each tool’s permissions and keep credentials separate. Start with read-only tasks to verify behavior.

Does Cursor require a remote server?

No. Cursor supports local stdio as well as SSE and Streamable HTTP transports. The available choice depends on the server.

Are these services unlimited because they are free to start?

No. Check each provider’s current rate, account and plan limits before relying on it for a large workload.

Can an MCP server take screenshots inside Cursor?

Yes. ScreenshotNeo’s MCP server includes screenshot, page-information and PDF capture tools, and its HTTP endpoint can be called from your own scripts.