MCP servers for Claude Code · GitHub & code hosting
Best GitHub & code hosting MCP server for Claude Code, tested
In ToolPassed's hands-on tests in Claude Code, DeepWiki MCP is our provisional candidate for GitHub & code hosting: it passed install, connect, tools and an agent task on 2026-10-22. We ran 3 GitHub & code hosting MCP servers in Claude Code; 3 passed every step.
Last tested · runs from 2026-10-15 to 2026-10-22 · one run per server · How we test
Results in Claude Code
| Server | Claude Code |
|---|---|
| DeepWiki MCP GitHub & code hosting | Passed 2026-10-22 · install, connect, tools, agent task |
| GitHub Repo MCP GitHub & code hosting | Passed 2026-10-21 · install, connect, tools, agent task |
| GitMCP GitHub & code hosting | Passed 2026-10-15 · install, connect, tools, agent task |
Dated results
- DeepWiki MCP passed in Claude Code on 2026-10-22 (Claude Code 2.1.296, mcp-remote 0.15.2, connected in 5.7s). Full result
- GitHub Repo MCP passed in Claude Code on 2026-10-21 (Claude Code 2.1.296, github-repo-mcp 1.0.3, connected in 5.6s). Full result
- GitMCP passed in Claude Code on 2026-10-15 (Claude Code 2.1.296, server reported v1.1.0, connect time not recorded). Full result
Why DeepWiki MCP is the candidate
DeepWiki MCP passed every step in Claude Code. It is the group’s site-wide candidate (judged 2026-10-22): Read the README of the public modelcontextprotocol/servers repo and returned its title in all 4 clients we test, with no token. GitMCP also passed all 4 but took 8 calls in Claude Code and Cursor versus DeepWiki's 3. GitHub Repo MCP (the earlier pick, 2 calls) passed twice in Claude Code and Cursor but failed in ChatGPT (Codex) on Oct 19, so by judging rule 1 it ranks below both. The official GitHub MCP server needs a personal access token, so it isn't tested and can't be picked. Confirmed Oct 22, 2026: second runs passed in all 4 clients (Claude Code, Cursor and ChatGPT (Codex) at ~1 PM ET, Claude Desktop at ~10:35 PM ET).
Provisional: based on one dated run per app (no repeat runs yet), so treat it as a candidate, not a final pick. How we test.
Same servers in all 4 apps
| Server | Claude Code | Cursor | Claude Desktop | ChatGPT (Codex) |
|---|---|---|---|---|
| DeepWiki MCP GitHub & code hosting | Passed 2026-10-22 · install, connect, tools, agent task | Passed 2026-10-22 · install, connect, tools, agent task | Passed 2026-10-22 · install, connect, tools, agent task | Passed 2026-10-22 · install, connect, tools, agent task |
| GitHub Repo MCP GitHub & code hosting | Passed 2026-10-21 · install, connect, tools, agent task | Passed 2026-10-21 · install, connect, tools, agent task | Passed 2026-10-18 · install, connect, tools, agent task | Failed 2026-10-19 · agent task failed (no successful call to this server's tools; answer missing ['Model Context Protocol servers']; no tool result containing ['Model Context Protocol servers']) |
| GitMCP GitHub & code hosting | Passed 2026-10-15 · install, connect, tools, agent task | Passed 2026-10-15 · install, connect, tools, agent task | Passed 2026-10-18 · install, connect, tools, agent task | Passed 2026-10-19 · install, connect, tools, agent task |
What failed or is partial in Claude Code
None of these servers failed in Claude Code.
Install in Claude Code
The exact snippet we tested is on the DeepWiki MCP page (pick the Claude Code tab). Not connecting? Claude Code MCP troubleshooting.
Not tested in Claude Code yet
GitHub MCP, GitLab MCP (zereight). Listed from their source; no result in Claude Code yet.
Questions
What is the best GitHub & code hosting MCP server for Claude Code?
In ToolPassed's hands-on tests in Claude Code, DeepWiki MCP is our provisional candidate for GitHub & code hosting: it passed install, connect, tools and an agent task on 2026-10-22. We ran 3 GitHub & code hosting MCP servers in Claude Code; 3 passed every step.
Does DeepWiki MCP work with Claude Code?
DeepWiki MCP passed in Claude Code on 2026-10-22 (Claude Code 2.1.296, mcp-remote 0.15.2, connected in 5.7s).
More: all MCP servers tested in Claude Code · GitHub & code hosting in every app · How we test