Flagship Collection · 2026

Which OpenClaw MCP servers to install first

We re-ranked the list around maintained projects, official support, current API drift, and real OpenClaw workflows. Stars are a signal, not the whole ranking. Pick the one task below that fits yours, copy that single minimal config, and check it in the config checker. You don't need all ten.

Last reviewed: 2026-05-25. Verify against the latest OpenClaw and MCP server docs before production use.

ONE TASK. ONE STARTING POINT.

What do you want to do first?

Choose one MCP, check the config, then verify it in your client. Add more once it works.

Choose your first task
1

Prepare Context7

Make sure Node.js and npx are available. The public endpoint is rate-limited; follow the official docs for API key setup if needed.

Minimal Context7 config json
{
  "mcpServers": {
    "context7": {
      "command": "npx",
      "args": ["-y", "@upstash/context7-mcp@latest"]
    }
  }
}

Paste into an MCP client configuration that supports this JSON format. If mcpServers already exists, merge this entry and keep your other settings. Check the setup guide for your OpenClaw integration. Setup guide →

2

Check this config

Check JSON syntax and common configuration issues. Passing a static check does not confirm installation.

Open config checker
3

Verify in your client

After saving the configuration and reloading MCP, try this prompt:

Use Context7 to find documentation for the library I am using. Give one example and identify the library version and documentation source.

Confirm that your client actually invokes Context7 and returns documentation for the intended library. Full setup and troubleshooting →

1

Prepare Filesystem

Create openclaw-workspace with one harmless test file. Relative paths resolve from the server’s working directory; use the folder’s absolute path if needed.

Minimal Filesystem config json
{
  "mcpServers": {
    "filesystem": {
      "command": "npx",
      "args": ["-y", "@modelcontextprotocol/server-filesystem", "./openclaw-workspace"]
    }
  }
}

Paste into an MCP client configuration that supports this JSON format. If mcpServers already exists, merge this entry and keep your other settings. Check the setup guide for your OpenClaw integration. Setup guide →

2

Check this config

Check JSON syntax and common configuration issues. Passing a static check does not confirm installation.

Open config checker
3

Verify in your client

After saving the configuration and reloading MCP, try this prompt:

List files in openclaw-workspace, then read the test file and summarize its contents.

Confirm that access stays within the folder you allowed. Do not expose your home directory or credential folders. Full setup and troubleshooting →

Explore the full Top 10 list ↓

Who this is for / not for

  • You have OpenClaw running and need a sane first MCP stack
  • You want to balance value, setup effort, and security
  • You need a way to narrow config failures before adding more servers

Avoid this

  • Installing every shiny MCP at once
  • Opening data access before deciding what AI may read
  • Bypassing OAuth, token scopes, or organization security controls

When existing MCPs should not own your private workflow

If your workflow needs fixed rules, private APIs, narrow permissions, or stable output, the next step is not installing more MCPs. Design a small reliable custom MCP.

Read the custom MCP builder guide

Ask these 3 questions before installing

Current combinations for specific roles.

  • Is it official or actively maintained, rather than just popular a few months ago?
  • Does it need a token or OAuth, and can that be handled through env vars or hosted auth?
  • Can it read or write local files, repos, or production data, and can that access be narrowed?
💻

Full-Stack Dev Kit

Context7 + GitHub + Filesystem + Playwright

📰

Research Kit

Firecrawl + Exa + Context7 + Fetch

🚀

Team Ops Kit

Slack + Memory + Filesystem + Linear

Common selection/config errors

Too many servers at once

Symptom: OpenClaw starts slowly or several servers fail together. Cause: a full stack was pasted before one server was verified. Fix: keep one server, then add the rest back one by one. Ask for review if you cannot isolate the failing block.

API key missing

Symptom: search, GitHub, or SaaS MCPs report missing credentials. Cause: placeholders were copied as real values or env vars were not injected. Fix: use environment variables; never send raw keys for review.

OAuth failed

Symptom: auth redirect fails or callback never completes. Cause: account, scopes, callback, or organization policy mismatch. Fix: verify with official docs, then send the sanitized error if still blocked.