Guides · Claude Desktop

How to add a remote MCP server to Claude Desktop and claude.ai

In Claude (claude.ai, Claude Desktop, Cowork) remote MCP servers are added as custom connectors. Claude connects to them from Anthropic's cloud, so the server must be reachable on the public internet.

Steps summarize the official docs linked below, checked 2026-10-19. Client UIs change; if a menu name differs, the linked docs win.

1. 1. Open connectors

   Go to Customize → Connectors. On Pro and Max: click + Add, then Add custom connector (some layouts show Add → Custom → Web). On Team and Enterprise, an Owner adds it under Organization settings → Connectors first.

2. 2. Enter name and URL

   Give it a name and paste the server's remote MCP URL, for example https://agenttaskatlas.com/api/mcp. Click Continue and review the authentication Claude detected.

3. 3. Choose authentication

   Sign in now or Sign in when needed for OAuth servers; No sign in for open servers like the AgentTask Atlas catalog. For an API key, choose No sign in and add the key under Request headers.

4. 4. Prefer the local config file?

   claude_desktop_config.json launches local stdio servers. To reach a remote URL from it you need a bridge such as the community mcp-remote package (not an Anthropic tool).

   {
     "mcpServers": {
       "agenttask-atlas": {
         "command": "npx",
         "args": [
           "-y",
           "mcp-remote",
           "https://agenttaskatlas.com/api/mcp"
         ]
       }
     }
   }

Good to know

• Free plans can add one custom connector.
• Custom connectors can reach services Anthropic has not verified; only connect servers from organizations you trust.
• claude_desktop_config.json lives in ~/Library/Application Support/Claude/ on macOS and %APPDATA%\Claude\ on Windows.

Sources

• Claude Help Center: Get started with custom connectors using remote MCP ↗
• Claude Help Center: Use connectors ↗

27 catalog entries with documented Claude Desktop setup

From the AgentTask Atlas catalog: entries whose official README or docs include setup instructions for Claude Desktop or Claude.ai.

• Filesystem · MCP steering group (reference) · no credentials
  Pick it when the agent must read/write project files; scope it to the directories you pass on the command line.
• Git · MCP steering group (reference) · no credentials
  Pick it for local Git work (status, diff, commit, branch) without a hosted account; it can modify the repo.
• Memory (knowledge graph) · MCP steering group (reference) · no credentials
  Pick it for persistent notes across sessions; it stores a local JSON knowledge graph, no cloud account.
• Sequential Thinking · MCP steering group (reference) · no credentials
  Pick it to make the agent plan step by step; no external access and no credentials.
• webapp-testing · Anthropic · no credentials
  Pick it to have the agent test a local web app with Playwright scripts; runs code on your machine.
• mcp-builder · Anthropic · no credentials
  Pick it when you are building your own MCP server; it guides and scaffolds code in your project.
• skill-creator · Anthropic · no credentials
  Pick it to write a new Agent Skill; it creates skill files in your project.
• frontend-design · Anthropic · no credentials
  Pick it for UI work; instructions only, so nothing runs beyond the code the agent writes.
• claude-api · Anthropic · needs credentials
  Pick it when building an app on the Claude API; the code you ship needs ANTHROPIC_API_KEY.
• web-artifacts-builder · Anthropic · no credentials
  Pick it for quick React/Tailwind prototypes; it runs local bundling scripts.
• doc-coauthoring · Anthropic · no credentials
  Pick it to co-write specs or docs; instructions only, edits the documents you provide.
• pdf · Anthropic · no credentials
  Pick it to read, fill or generate PDFs; runs local Python scripts on your files.
• GitHub MCP Server · GitHub · needs credentials
  Pick it for issues, PRs and Actions from the agent; limit toolsets or use read-only mode for review tasks.
• Playwright MCP · Microsoft · no credentials
  Pick it to let the agent drive a real browser for E2E checks; it can reach any site the browser can.
• AWS MCP Servers · AWS Labs · needs credentials
  Pick the specific AWS server you need; the docs server needs no credentials, others use your IAM profile.
• Terraform MCP Server · HashiCorp · credentials optional
  Pick it for Terraform Registry lookups; add TFE_TOKEN only if you need HCP Terraform workspaces.
• Grafana MCP · Grafana Labs · needs credentials
  Pick it to query dashboards, Prometheus and Loki from the agent; needs a Grafana service account token.
• Notion MCP Server · Notion · needs credentials
  Pick it to read/write Notion specs; it only sees pages shared with the integration.
• Docker MCP Gateway · Docker · credentials optional
  Pick it to run several MCP servers in containers; permissions depend on the servers you enable.
• Firecrawl MCP Server · Firecrawl · needs credentials
  Pick it for crawling and scraping at scale; URLs are sent to Firecrawl's hosted API with your key.
• Elasticsearch MCP Server (deprecated) · Elastic · needs credentials
  Avoid for new setups: the official README marks it deprecated; use it only on existing installs.
• Heroku Platform MCP Server · Heroku · needs credentials
  Pick it to manage Heroku apps from the agent; the README recommends `heroku mcp:start` so you don't expose an API key.
• Atlassian Rovo MCP Server · Atlassian · needs credentials
  Pick it when the agent must read or update Jira issues, Confluence pages or Bitbucket repos; the README recommends the v2 endpoint for new setups.
• Redis MCP Server · Redis · needs credentials
  Pick it to let the agent inspect or cache data in Redis; connect it with a read-only ACL user unless writes are intended.
• Microsoft Learn MCP Server · Microsoft · no credentials
  Pick it when the agent writes Azure, .NET or other Microsoft code and should look up current official docs and samples; it needs no credentials.
• Buildkite MCP Server · Buildkite · needs credentials
  Pick it to let the agent investigate failed Buildkite builds and logs; use the read-only endpoint unless you want it to retry jobs or start builds.
• Datadog MCP Server · Datadog · needs credentials
  Pick it to investigate production issues from the agent using your Datadog account; OAuth keeps access within your existing RBAC.

Ask the catalog from inside Claude Desktop: the AgentTask Atlas MCP server is free, read-only and needs no account.

Compare by task: Databases & SQL · Infra & deploy · Debugging & observability · Library docs & research · Frontend & UI · Docs & writing

⚠ No independent security audit: entries summarize official sources only. Review permissions and code before installing.