switchboard-agents
v0.1.0
Published
Coordinate AI coding agents (Claude Code, Cursor) through a shared FalkorDB import graph: claims, conflict checks, memory, an MCP server and editor hooks.
Maintainers
Readme
switchboard-agents
Switchboard stops AI coding agents from breaking each other's work. It keeps your repo's import graph in FalkorDB. Every edit an agent makes is claimed in that graph. An edit within 3 import hops of a file another agent holds is refused, with the import path, the holder and the files that are safe now.
Site and live demo: https://switchboard-agents.xyz · Source: https://github.com/bhargava-gumpula/graph-hacks
Needs
- Node 22 or later
- FalkorDB:
docker run -d -p 6379:6379 falkordb/falkordb(or setFALKORDB_URL)
Use it
npm i -g switchboard-agents
cd your-repo
switchboard init # Cursor and Claude Code hooks + MCP config, then parses the repo
switchboard init --cursor # Cursor onlyRestart the editor in that repo. Every edit is now claimed for you. Nothing to call.
switchboard mcp(or theswitchboard-mcpbin) is the stdio MCP server, 15 tools.switchboard status,check <file>,brief --prefix "",why <file>,blast <file>and more:switchboard --help.
Claude Code users can install the plugin instead, which bundles the same hooks, the MCP server and a skill:
/plugin marketplace add bhargava-gumpula/graph-hacks
/plugin install switchboard@switchboardSettings
| Variable | Default |
| --- | --- |
| FALKORDB_URL | redis://localhost:6379 |
| SB_GRAPH | sb_<repo name>, shared by every worktree of the repo |
| SB_AGENT | <editor>@<branch>, such as cursor@main |
| SB_URL | none. A Switchboard dashboard to link in the session brief |
| SWITCHBOARD_OFF | set to 1 to let every edit through |
MIT License.
