Suvadu vs McFly: Shell History Features Compared (2026)
Searching for a McFly alternative? Suvadu vs McFly compared on search, ranking, privacy, AI agent support, and what each one is missing.
McFly is a smart, lightweight upgrade to Ctrl-R — it replaces your default reverse search with a neural-network-ranked one that prioritises commands you're likely to want. It's a good tool. But it's also deliberately narrow: it's a better search box, and not much more. If you're looking for a McFly alternative that does more than rank Ctrl-R results, here's how Suvadu compares.
Updated for Suvadu 0.5.1, which added frecency, failure-rate and next-command signals to Suvadu's ranking; the ranking rows below changed accordingly.
Suvadu vs McFly at a glance
| Feature | Suvadu | McFly |
|---|---|---|
| Language | Rust | Rust |
| Storage | SQLite (WAL mode) | SQLite |
| Core idea | Full history workspace | Smarter Ctrl-R |
| Search ranking | Literal word/substring matches, then a rule-based Smart rank (directory, frecency, failure rate, what you usually run next) + filters | Neural ranking |
| No exact match | Retries with separators ignored (hiveagent finds hive-agent), labelled as such | Fuzzy matching |
| History interface | Full-screen TUI with filter panel | Ctrl-R search interface |
| Storage / sync | Local database; no built-in cloud sync | Local database |
| AI agent tracking | Claude Code, Codex, Cursor, OpenCode, pi.dev; Antigravity and others by detection | — |
| MCP server | Built-in (22 read-only tools) | — |
| Risk assessment | Yes | — |
| Stats / heatmap | Yes | — |
| Bookmarks & notes | Yes | — |
| Cross-machine export | Plaintext JSONL/JSON/CSV; JSONL import | — |
Where McFly is great
- It's focused. McFly does one thing — ranked history search — and does it well. The neural ranking genuinely surfaces the right command more often than a plain reverse search.
- It's lightweight. If all you want is a better
Ctrl-Rand nothing else, McFly stays out of your way. - It's local. Like Suvadu, McFly keeps everything on your machine. No accounts, no servers.
If your only complaint with shell history is that reverse search is dumb, McFly fixes that and you can stop reading here. The case for Suvadu is for everyone who wants more than search.
Ranking: closer than it was
Until 0.5.1, Suvadu ranked typed queries by how well the text matched, with a boost for the current directory, and did not weigh how often you ran a command or whether it worked. That was the clearest gap against McFly. Suvadu 0.5.1's Smart rank adds three signals within a match tier: frecency (each run counts 16 in its first hour, falling to 0.25 after a month, so twenty runs a day or two ago beat one run five minutes ago), an outcome penalty for a command that fails more than half of at least three runs, and a sequence boost for what you usually run after this shell's last command — after git add -A, git commit comes first.
The approaches still differ. McFly trains a small neural network on your own choices. Suvadu's rules are fixed and written down in its search docs, a closer textual match always beats any boost, and Ctrl+S switches to a Recent rank with none of them. Neither project publishes a ranking-quality comparison, and Suvadu's own benchmark notes that ranking quality is not yet measured, so which ranks better for you is something to try rather than read.
1. History as a workspace, not just a search box
McFly improves the moment you press Ctrl-R. Suvadu treats your whole history as something you work with: a full-screen search TUI with a filter panel (by directory, time, exit code, session, executor), recency-first arrow-key history, bookmarks, notes, tags, and a stats dashboard with a heatmap.
2. It knows what your AI agents ran
McFly does not add an executor field to its history records. Suvadu records an executor for commands captured through its shell and agent integrations; prompt text is attached only when the selected integration captures it. For Claude Code, Codex, OpenCode and, since 0.5.1, Cursor, it also captures the agent's session and which files that session changed in git. You can review, search, and replay that captured activity and flag commands with rule-based risk assessment. See agent tracking.
3. Your history becomes agent memory
Suvadu's local MCP server lets a connected AI client query captured history. Returned data is available to the client and may be sent to its provider under the client's settings. McFly does not include an MCP server.
Performance
Both tools are implemented in Rust, but this article does not include a reproducible head-to-head benchmark. Evaluate responsiveness with your own history size and shell configuration rather than treating the title as a performance result.
So, McFly or Suvadu?
- Stick with McFly if you want a minimal, neural-ranked replacement for
Ctrl-Rand nothing more. - Move to Suvadu if you want a full history workspace, you work with AI coding agents and want to track what they run, or you want your history available to your agent through MCP.
Installing Suvadu provides the binary; shell recording starts after you add the Zsh or Bash initialization line. The importer accepts Suvadu JSONL exports, raw Zsh and Bash history files, and an Atuin history database.
Install Suvadu → · Compare Atuin, fzf, McFly, hstr and Suvadu →
Builder of Suvadu. Writes Rust, thinks about shell history more than most people, and believes developer tools should be local-first.