llm-collab
This server does not currently have a deployable release on Glama. Follow the repository's setup instructions to install and configure it for your MCP client, including any required credentials and local dependencies.
llm-collab
Local coordination for Claude Code, Google Antigravity and OpenAI Codex. All three MCP clients can share a task ID, recover a handoff, and inspect a decision. The coordinator can run a bounded discussion, assign implementation to separate Git checkouts, and preserve progress when a provider reports a quota failure.
This project connects to an independently installed local-agent-bridge. Compatibility was verified with Bridge commit c0f9486ecce2570bb14a806a238f12201621df14. Install and configure Bridge and the provider CLIs using their respective instructions.
Configure
Use Node.js 22 or later and Git on macOS or Linux. Clone this repository, then install its locked dependencies and copy the configuration template.
npm ci --ignore-scripts
cp config.example.json config.json
node bin/llm-collab.mjs --config config.json --checkEdit config.json before running the check. Relative paths are resolved from the directory containing that file.
Setting | Purpose |
| External Bridge installation with its own dependencies installed |
| Standalone Git checkout used for local collaboration. It needs an |
| Writable local task, quota and handoff state. Use a short absolute path if the platform limits Unix socket path length |
| CLI names on PATH or explicit paths for Claude Code, Antigravity and Codex |
| Optional deployment root for reading |
Register the same stdio command in all three clients. Use the absolute Node.js, launcher and config paths in client configurations.
codex mcp add llm-collab -- /path/to/node /path/to/llm-collab/bin/llm-collab.mjs --config /path/to/config.json
claude mcp add --scope user --transport stdio llm-collab -- /path/to/node /path/to/llm-collab/bin/llm-collab.mjs --config /path/to/config.json
agy mcp add llm-collab -- /path/to/node /path/to/llm-collab/bin/llm-collab.mjs --config /path/to/config.jsonRelated MCP server: agent-bridge
Work together
Use
workspace_statusfor the active checkout andteam_statusfor quota observations and coordinator priority. Among providers with confirmed availability, the order is Claude Code, Antigravity, then Codex.Use
start_collaborationin review mode for discussion. For an implementation phase, use work mode with a writer and an optionaltaskBranchfor a private checkout. Give independent tasks different branch names.Use
list_collaborationsandget_collaborationfrom any client to find a task and its latest handoff.continue_collaborationuses that same task ID.A reversible technical decision can be put to a vote. Each participant emits a line such as
VOTE: {"choice":"A","reason":"The benchmark is reproducible."}.tally_team_voterecords a decision when one option gets at least two votes and a majority. The receipt includes dissent and a rollback route.When a known usage window reaches 3% remaining, the launcher blocks additional delegated turns for that provider. At a safe stopping point, use
save_team_handoffto save the summary, next step, and individual subagent notes. Existing interactive sessions follow their own workspace’s checkpoint instructions. A Bridge installation with compatible quota collectors can report Codex and Claude percentages. Antigravity reaches the fallback path on confirmed provider failure.
The repository ships the coordinator and config template. The external Bridge checkout, credentials, local state, transcripts, and development workspaces are installed or created on each host. Keep local configuration and runtime data outside commits. Publishing and production operations run through the host’s established authorization and deployment process.
Run npm test for local vote and handoff checks.
Tool Schema Changelog
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
Loading tool schema changelog…
This server cannot be deployed
Maintenance
Related MCP Connectors
Git collaboration for AI agents and people: tasks, hand-offs, reviews, teams, shared memory.
Shared task layer for AI agents: tasks, handoffs and decisions that persist across tools.
One shared project for a team's coding agents: decisions, a task board and handoffs.
Shared long-term memory & Stop-hook task handoff across Claude Code, Codex, Cursor, and mobile.
Related MCP Servers
- AlicenseNot gradedqualityCmaintenanceCoordinates parallel AI coding agents by providing task ownership, scoped file locks, handoffs, and verification workflows.MIT
- AlicenseNot gradedqualityAmaintenanceEnables coordinating Claude Code and Codex across separate Git worktrees with shared issue ownership, file reservations, messages, and explicit handoffs.213 PyPI3MIT
- AlicenseNot gradedqualityAmaintenanceEnables AI coding providers to run structured multi-agent workflows with document specifications, tracked tasks, handoffs, and server-owned health evidence. It coordinates task claims, repairs, reviews, and completion gates across Claude Code, Codex CLI, Cursor, Grok Build, and OpenCode.29 npmMIT
- AlicenseNot gradedqualityAmaintenanceEnables AI coding providers to run structured multi-agent workflows with tracked tasks, document specifications, handoffs, health gates, and server-owned evidence across providers like Claude Code, Codex CLI, Cursor, Grok Build, and OpenCode.29 npmMIT