MCP servers

Adding your own MCP server

aiflow generates one MCP config per enabled agent (.aiflow/config.json → agents.*) from the same server set — .mcp.json for Claude Code, .codex/config.toml for OpenAI Codex CLI, .vscode/mcp.json for GitHub Copilot in VS Code. The Model Context Protocol (MCP) servers it can wire:

Server	Purpose	Enabled by
git-host (github / gitlab / bitbucket / gitea / forgejo)	issues, PRs/MRs on your host	`remote.mcp`
GitKraken	workspaces/PRs/issues via the `gk` CLI — a client, wired alongside any host	`gitkraken.enabled`
filesystem	safe structured file access	`mcp.filesystem`
graphify	structural code knowledge graph (imports/call-graph)	`graphify.enabled`
cocoindex-code	semantic code RAG search (`ccc mcp`)	`mcp.cocoindex`
context7	live, version-correct library docs	`mcp.context7`
task-master	goal/PRD → task decomposition	`taskmaster.enabled`

The git-host MCP is chosen per remote.type, with the base URL threaded in (GITHUB_HOST / GITLAB_API_URL / GITEA_URL) so enterprise/self-managed hosts work. Tokens come from .env (remote.tokenEnv), referenced as ${VAR} placeholders — Codex CLI/VS Code may expect a different interpolation syntax; see Multi-Agent Support for the caveats per agent.

Adding your own MCP server

Add an entry to .mcp.json — servers aiflow doesn’t manage are preserved on re-render:

{
  "mcpServers": {
    "my-server": {
      "command": "npx",
      "args": ["-y", "@scope/my-mcp-server"],
      "env": { "MY_TOKEN": "${MY_TOKEN}" }   // secrets via .env, never inline
    }
  }
}

Then allow it in .claude/settings.json under permissions.allow (e.g. "mcp__my-server") and put the secret in .env. Browse vetted servers with npx claude-code-templates@latest. Prefer a focused MCP over a broad one — fewer tools = less context and fewer wrong turns.

See Configuration and the MCP servers ecosystem.