collidemcpBLOCK
Run multiple AI coding agents (Claude Code, Codex, Cursor) on one repo without conflicts: live awareness, write gate, briefings, one-step pushes. Free, local, MCP + hooks.
Overview
From the repository's own README, as read at the audited commit. Badges and raw HTML are left out.
[](#what-is-in-this-repository) [](https://www.npmjs.com/package/collidemcp) [](https://collidemcp.com/docs) [](https://collidemcp.com)
Collide is the coordination layer for multiplayer coding agents: it keeps Claude Code, Codex and Cursor agents, on one machine or across a team's machines, from overwriting each other on one codebase, and hands each agent what the others already learned.
This repository is Collide Free: run as many coding agents as you like on one machine, in as many repositories as you like, without them tripping over each other. Free, no account, and nothing leaves your machine.
curl -fsSL https://collidemcp.com/install.sh | sh
or, with Node (any terminal, including Windows PowerShell):
npx collidemcp
One install covers every repository on the machine. Agents already open pick it up at their next prompt.
Why
Run two or three Claude Code agents (or Codex, or Cursor) on the same codebase and they start colliding: one renames a function another is still calling, two write the same module, a push fails and the agent burns turns pulling and re-testing. Each agent also re-reads the same files the others just read. Separate git worktrees keep their files apart, but the work still meets at merge time.
Collide sits under the agents you already run. It does not run agents or host machines. Its hooks tell each agent what the others are changing while they work, hold a write that would clash, and hand each prompt the exact code it is about, so agents search less.
In Collide's benchmark (12 Claude Code agents at once on one repository, hard
d582fc659be4OBSERVED · 2026-10-11Connect
Built from this server's own package name, version and transport as found in its source — not copied from anyone's documentation, so it cannot drift against a page we do not control.
claude mcp add collide-core -- uvx collide-core
{
"mcpServers": {
"collide-core": {
"command": "uvx",
"args": [
"collide-core"
]
}
}
}Trust audit
BLOCKgrade F · trust 52/100 Do not install this without reading the findings. The audit found something that could harm you or your machine.
| Layer | What it checks | Result |
|---|---|---|
| L0 | Provenance & inventory | WARN |
| L1 | Static analysis of the code | FAIL |
| L2 | Instruction surface (what it tells the agent) | PASS |
| L3 | Class-specific surface | WARN |
| L4 | Behavioural (sandbox) | SKIPPED |
What the source does
- Filesystem
- declared (9 observation(s))
- Network
- declared (10 observation(s))
- Shell
- none-observed
- Dependencies
- pinned
- Secrets in source
- found
Findings (18)
("curl -H \"Authorization: Bearer ghp_0123456789abcdefghijABCDEFGHIJ\" https://x", "curl -H \"Authorization: Bearer [redacted]\" https://x"),("STRIPE_KEY=sk_live_abcdef1234567890 npm run seed", "STRIPE_KEY=[redacted] npm run seed"),assert!(message_rewrite(&input("collide message \"$(cat ~/.ssh/id_rsa)\"")).is_none());"id_rsa", "token",
collide.icns
std::env::set_var("COLLIDE_PUBLIC_URL", format!("http://127.0.0.1:{port}"));let redirect = format!("http://127.0.0.1:{port}/callback");.send_json(json!({"client_name": client_name, "redirect_uris": ["http://127.0.0.1/callback"],let value = "http://127.0.0.1:5555/callback?x=a b&y=é";
(format!("http://127.0.0.1:{port}"), token)streamable-http
include_str!("../../client/src/collide_client/report_hook.py")include_str!("../../client/src/collide_client/gate_hook.py")include_str!("../../client/src/collide_client/pre_commit.py")assert!(!safe(".collide/../../x"));curl -fsSL https://collidemcp.com/install.sh | sh
This plugin is for Claude Code. For Codex and Cursor too, install Collide for the whole machine: `curl -fsSL https://collidemcp.com/install.sh | sh` or `npx collidemcp` ([collidemcp](https://github.co
Gates applied: critical_finding, no_behavioural_pass.
d582fc659be4full audit observations/trust-audit/mcp-server/lithometric__collidemcp.json · Report an issue / request a re-scanAudit history
Every audit this server has had. A grade with a past is a grade somebody is still checking.
| Date | Source | Verdict | Grade | Score | Change |
|---|---|---|---|---|---|
| 2026-10-11 | d582fc659be4 | BLOCK | F | 52 | first audit |
Questions
What is the collidemcp MCP server?
Run multiple AI coding agents (Claude Code, Codex, Cursor) on one repo without conflicts: live awareness, write gate, briefings, one-step pushes. Free, local, MCP + hooks.
Is collidemcp safe to connect to an agent?
No — not without reading the findings first. The audit graded it F (52/100) and found 4 critical or high issues in the source. Each one is listed on this page with the file and line it is on.
What credentials does collidemcp need?
No credential environment variables were found in its source, so it appears to need none.
How does collidemcp run?
It speaks streamable-http, so it runs as a service you connect to over the network. It is published on PyPI as collide-core.
How current is this page?
The grade is for one exact copy of the source (d582fc659be4), read on 2026-10-11. The repository is watched and re-audited when it changes.