Claude Code and Codex are fast, file-editing CLIs, but their base installs are quiet on purpose. They can read files, edit files, and run shell commands—and that’s it. The moment you want either one to pull real docs, hit a real browser, or check a real deployment, you need MCP servers.
Firecrawl’s Rhys Hanak, who switches between the two CLIs daily, put together a tried-and-tested list of the best MCP servers for Claude Code and Codex in 2026. The post is CLI-first: every install command is written twice, once for claude mcp add and once for codex mcp add, with exact JSON and TOML config snippets.
What MCP actually solves
MCP (Model Context Protocol) is an open standard from Anthropic that lets an AI client talk to external tools and data sources through one shared interface. Before MCP, every combination of AI client and tool needed its own integration—GitHub had to be wired separately for Claude, Cursor, ChatGPT, and VS Code. Anthropic called this the “N x M problem.” MCP replaces all of that with a single JSON-RPC 2.0 protocol.
Since Anthropic released MCP in November 2024, the ecosystem has grown to over 17,000 publicly listed servers on mcp.so. OpenAI adopted MCP for Codex in early 2025, Google DeepMind followed, and the standard was donated to the Linux Foundation’s Agentic AI Foundation in December 2025. In Zuplo’s State of MCP survey, 72% of users said their MCP usage will grow in the next 12 months, 70% already run 2 to 7 servers at once, and 49% cited developer productivity as the primary ROI.
The config difference between Claude Code and Codex
Both CLIs speak the two transports that cover 99% of servers: stdio (local subprocess) and HTTP (remote, usually with OAuth). Claude Code additionally supports SSE and WebSocket; Codex sticks to stdio and HTTP.
Config file locations differ. Claude Code uses JSON under a top-level mcpServers object in ~/.claude.json (user scope) or .mcp.json (project scope). Codex uses TOML under [mcp_servers.<name>] tables in ~/.codex/config.toml. Hanak notes two gotchas: in Claude Code, a remote server entry with a url but no type is a config error, and Codex silently skips misnamed sections—the table name must be mcp_servers, not mcp.servers.
You rarely need to edit files by hand. Each CLI writes the config for you:
# Claude Code, remote HTTP with OAuth
claude mcp add --scope user --transport http firecrawl https://mcp.firecrawl.dev/v2/mcp-oauth
# Codex, remote HTTP with OAuth
codex mcp add firecrawl --url https://mcp.firecrawl.dev/v2/mcp-oauth
codex mcp login firecrawlCodex splits registration and authentication: codex mcp add writes the config, then codex mcp login <name> opens the browser to complete OAuth. Claude Code triggers OAuth inline on first use. Inside either CLI, /mcp shows every configured server, whether it’s connected, and which tools it exposes.
The server that stands out: Firecrawl MCP
Hanak’s top pick is Firecrawl MCP, which exposes Firecrawl’s web context stack—Search, Scrape, Parse, Crawl, Map, Interact, Agent, and Monitor—so both Claude Code and Codex can call every endpoint directly. Neither CLI has a reliable way to hit the live web out of the box; shelling out to curl returns raw HTML, gets blocked on most modern sites, and cannot render JavaScript.
The tool that makes this MCP different for coding agents is firecrawl_developer_search. It queries the Firecrawl Developer Index, a semantic index over 70M+ developer artifacts refreshed daily: READMEs, GitHub issues, pull requests, OpenAPI specs, changelogs, and external docs sites. The index ranked #1 overall on the DevDex benchmark, which scores how often a tool returns the artifact an agent actually needs. In practice, that means fewer hallucinated APIs and answers that cite a real PR or issue instead of a plausible-sounding paragraph.
Install is straightforward, with both keyless and OAuth options:
# Claude Code, keyless (Search / Scrape / Parse only, rate-limited per IP)
claude mcp add --scope user --transport http firecrawl https://mcp.firecrawl.dev/v2/mcp
# Codex, keyless
codex mcp add firecrawl --url https://mcp.firecrawl.dev/v2/mcpHanak’s honest take: “This is the one MCP I refuse to work without in either Claude Code or Codex.” The keyless endpoint means a teammate can be searching the live web thirty seconds after receiving a Firecrawl URL. The main con is that large-scale web data extraction burns through Firecrawl credits fast, so serious crawl or agent workloads need a higher-tier plan.
The rest of the list, briefly
The full post covers 12 servers, each with a specific job neither CLI can do out of the box. Highlights include Playwright MCP for browser automation via the accessibility tree, Chrome DevTools MCP for console and network traces on a running app, GitHub MCP for PRs and cross-repo code search, Linear MCP for tickets and cycles from the terminal, Vercel MCP for build logs and deploys, Sentry MCP for full error context, and Sequential Thinking MCP for structured, revisable reasoning on hard problems.
If two servers cover the same job, Hanak kept the one with the cleaner surface and the more permissive auth story. Every install block shows the command for both CLIs, so you can wire the same open protocol into whichever agent you’re using that day.
Sources
AI-assisted summary compiled from the sources above, reviewed by a human before publishing.