s//slipstream
file session coordinator

Three MCP tools that give your agent self-contained batch edits, persistent file sessions, and conflict-safe writes. One daemon. Zero dependencies.

slipstream in action
01 GET STARTED

Two commands to three tools

Install the binary, register it as an MCP server, restart your client. That's it.

install
# macOS (Apple Silicon + Intel) and Linux $ curl -fsSL https://slipstream.sh/install.sh | sh # or from source $ cargo install --git https://github.com/aetherwing-io/slipstream
connect as MCP server
# Claude Code $ claude mcp add ss -- slipstream mcp # Cursor / Windsurf / any MCP client — add to config: "ss": { "command": "slipstream", "args": ["mcp"] }

Restart your client. Three tools appear:

ToolWhat it does
ss Edit files. ss(path, old_str, new_str) for one edit. ss(ops=[...]) for many. Self-contained — no setup needed.
ss_session Read files and manage sessions. ss_session("read <path>") to read. Run ss_help for advanced session control.
ss_help Full reference card — ops format, session control, advanced options.

macOS + Linux · Works with Claude Code, Cursor, Windsurf, or any MCP-compatible client

02 CAPABILITIES

What one daemon gives your agent

Quick Mode

One edit, one call

ss(path, old_str, new_str) — auto opens, replaces, flushes, closes. Same ergonomics as native Edit, but through slipstream.

Batch Mode

51 edits, still one call

ss(ops=[...]) handles any mix of str_replace, write, and read ops across unlimited files. Self-contained — no session setup needed.

Sessions

Named concurrent sessions

Multiple agents edit different files simultaneously without conflicts. Each session tracks its own buffer state independently.

Conflict Detection

External changes caught

External modifications detected before flush. Force-flush available when you know what you're doing.

Atomic Flush

All or nothing

Edits accumulate in memory, hit disk only on flush. Streaming writes with content hashing — no full-file allocation.

Compact Output

60% fewer tokens

FCP-style prefix lines (~ edit, > flush, @ read, ! error) plus a status bar. No JSON walls. Same information, fraction of the cost.

03 THE RECEIPTS

Measured, not promised

One agent, one task, four tool strategies. 20 Python files × 3 cross-cutting edits = 60 assertions. Every strategy scored 60/60 on correctness — the difference is cost.

StrategyTool CallsTokensWall TimeCorrect
SS + mish 6 16.7K 57s 60/60
SS MCP (batch) 7 19.5K 90s 60/60
Traditional
sed via Bash
7 14.9K 56s 60/60

51 file operations (20 header inserts + 20 logging adds + 11 renames) across 20 files. Run 4: Opus 4.6 subagents. The Traditional agent discovered sed batching on its own — fewer tokens, but no conflict detection, no sessions, no undo.