Geond Agent Protocol is a local-first MCP server for sharing memory, reserving files, handing off tasks, and reviewing evidence among multiple AI coding agents. It is aimed at agents collaborating on the same code repository, giving Copilot Chat, Codex, Claude Code, Antigravity, Manus, CLI agents, and MCP-enabled tools a persistent place to record: what happened, why it happened, what was changed, what was reserved, how it was verified, and what the next agent or reviewer should do.
Main capabilities
- Memory: importers normalize sessions, events, messages, usage records, and task history from agent tools, and redact common secrets before persistence.
- Code graph: Python, TypeScript, and JavaScript indexers connect files, symbols, imports, calls, references, and changesets.
- Reservations: agents can claim files or symbols with a TTL, with policy checks, renewal, release, and auditable reservation events.
- Handoff: agents leave structured next-action packages containing tested commands, blockers, remaining risks, and evidence references.
- Dashboard: humans and PM/orchestration agents can read compact overview, activity, timeline, code risk, usage, lineage, reservation, and handoff read models.
- Shared PostgreSQL: the local-first setup uses Docker PostgreSQL; team configurations can point local processes at Azure or other PostgreSQL-compatible backends.
Usage Prerequisites: Python 3.11+, uv, Docker with Compose, Git, and ripgrep. OS-specific instructions are in docs/developer_setup.md. Commands for starting the MCP server, previewing or writing MCP client configuration, and serving a read-only dashboard are given in the official documentation. You can run the stdio MCP server directly or from a Docker image. GEOND_DATABASE_URL points Geond at PostgreSQL; local development typically uses the Compose database from the quick start. Team mode can use GEOND_DATABASE_PROFILE=azure plus AZURE_GEOND_DATABASE_URL to share memory across machines while keeping each MCP process running locally.
Three representative MCP workflows
- Start coordinated work: review_workspace_context -> reserve_files -> edit -> record_changeset.
- Explain prior context: search_dev_memory -> explain_change -> get_changeset_detail.
- Leave a handoff: record_agent_action -> record_handoff_summary -> list_handoff_summaries.
Applicable scenarios Suitable for development scenarios where multiple AI coding agents collaborate on the same repository, need to share context, and avoid conflicts; also suitable for teams that want to share agent memory across machines while keeping processes local.
Limitations and caveats Geond is alpha software. Enterprise IAM, row-level security, a dedicated MCP audit stream, broad SaaS adapters, and automatic reservation of dependency extensions are on the roadmap and not yet implemented. Agent transcripts may contain sensitive information; review .env, transcripts, screenshots, benchmark logs, and dashboard screenshots before sharing. Do not commit private transcripts, local evidence exports, local drafts, repo, tmp, result, results, or generated videos. The project uses the Apache-2.0 license.
