# Concord AI

> The MCP collaboration layer for coding agents. A local MCP server and CLI that coding agents use while they work. Presence, task memory, ownership, agent-to-agent messaging, and review evidence in one shared workspace.

Concord AI is a local MCP server and CLI that coding agents call while they
work. MCP is the Model Context Protocol, an open standard for connecting coding
agents to external tools and data.

## Tools

- `start_work`: Register presence, claim or accept one task, and surface scope overlaps before editing.
- `inspect_work`: Read workspace or task state, an agent's inbox and outbox, or a prompt/reply thread.
- `update_work`: Append typed intent, decisions, assumptions, and findings, or prompt another agent live.
- `transfer_work`: Assign, accept, decline, release, reassign, offer a handoff, or reopen versioned work.
- `finish_work`: Record changes, tests, risks, and decisions, then mark the task review-ready before the PR.

## Links

- [Home](https://getconcord.ai)
- [Guides](https://getconcord.ai/guides)
- [Compare](https://getconcord.ai/compare)
- [Changelog](https://getconcord.ai/changelog)
- [Blog](https://getconcord.ai/blog)
- [Source](https://github.com/Get-Concord-AI/concord-mcp)
- [Feed](https://getconcord.ai/blog/feed.xml)

## Guides

- [Stop AI coding agents overwriting each other's files](https://getconcord.ai/guides/stop-ai-coding-agents-overwriting-each-other): Running two or three agents on one repo? Have each one claim files before editing, so the second agent learns about the first before it writes anything.
- [Let Claude Code and Codex talk to each other](https://getconcord.ai/guides/claude-code-codex-talk-to-each-other): Send direct, replyable messages between Claude Code, Codex, Cursor and other harnesses, without copying context between windows yourself.
- [Keep agent state after context compaction](https://getconcord.ai/guides/keep-agent-state-after-context-compaction): When a coordinator or long-running agent compacts its context, task status and decisions shouldn't go with it. Keep them in the workspace, not the chat.
- [Run many parallel coding agents without losing track](https://getconcord.ai/guides/run-many-parallel-coding-agents): Four agents plus subagents, or nine sessions at once: keep every lane's owner, scope and status visible from one place.
- [Worktrees don't isolate your database or ports](https://getconcord.ai/guides/worktrees-shared-database-ports-agents): Git worktrees keep agents' files apart, but they still share the dev database, migrations, ports and other local services. Declare those as shared risks.
- [Persist agent decisions between sessions](https://getconcord.ai/guides/persist-agent-decisions-between-sessions): Where does an agent's memory live between runs? Keep decisions, assumptions and "done" status attached to the task, where the next session can read them.
- [Block an agent from editing a file another agent claimed](https://getconcord.ai/guides/block-edits-to-claimed-files): Turn claims into a hard stop: a Claude Code PreToolUse hook, a git pre-commit hook or CI can refuse edits to files another active task owns.
- [Use git worktrees with parallel AI coding agents](https://getconcord.ai/guides/git-worktrees-for-ai-coding-agents): Worktrees give each agent its own checkout and branch. Here is what they solve, what they leave open, and how to keep worktree agents aware of each other.
- [Claude Code subagents vs parallel sessions](https://getconcord.ai/guides/claude-code-subagents-vs-parallel-sessions): Subagents work for one session and report back. Parallel sessions work independently. Each needs different coordination.
- [Use Claude Code and Cursor on the same project](https://getconcord.ai/guides/claude-code-cursor-same-project): Run Claude Code in the terminal and Cursor in the editor on one repository without each tool undoing the other's work.
- [Run Gemini CLI alongside Claude Code and Codex](https://getconcord.ai/guides/gemini-cli-with-claude-code-and-codex): Add Gemini CLI to a repo where Claude Code or Codex is already working, with shared claims, messages and task memory across all three.
- [AGENTS.md vs CLAUDE.md when you run several agents](https://getconcord.ai/guides/agents-md-claude-md-for-multiple-agents): Codex reads AGENTS.md, Claude Code reads CLAUDE.md, Cursor reads its rules. Keep standing instructions in one place, and keep live work-state out of all of them.
- [Add an MCP server for multi-agent coordination](https://getconcord.ai/guides/add-mcp-server-for-agent-coordination): One command registers a local coordination MCP server with Claude Code, Codex, Cursor, Gemini CLI and Grok Build in a repository.
- [Find work an AI coding agent abandoned mid-task](https://getconcord.ai/guides/find-abandoned-ai-agent-work): Sessions crash, get closed or hit limits. Spot claims whose agent has gone, see what it recorded, and hand the work to someone else.
- [Assign tasks to AI coding agents and know they accepted](https://getconcord.ai/guides/assign-tasks-to-ai-coding-agents): Give a task to a specific agent and get an explicit accept or decline, instead of assuming a prompt was read.
- [Do you need an orchestrator for multiple coding agents?](https://getconcord.ai/guides/orchestrator-vs-coordination-for-coding-agents): Orchestrators decide what runs. Coordination keeps agents that are already running from conflicting. Most teams with several coding agents need the second first.
- [See what all your coding agents are doing right now](https://getconcord.ai/guides/see-what-coding-agents-are-doing): One view of every agent in the repo: who is present, what each has claimed, live overlaps, stale claims and work ready for review.
- [Run multiple Claude Code agents in the same repo](https://getconcord.ai/guides/multiple-claude-code-agents-same-repo): Split work between parallel Claude Code sessions without duplicate edits, hidden assumptions, or agents colliding on the same task.
- [Prevent AI coding agents from conflicting](https://getconcord.ai/guides/prevent-ai-coding-agent-conflicts): Stop two coding agents from editing overlapping work or making incompatible decisions before the conflict reaches review.
- [Share context between Claude Code, Codex and Cursor](https://getconcord.ai/guides/share-context-between-coding-agents): Give different coding agents the same live understanding of tasks, decisions, blockers, and work already in progress.
- [Hand off work between AI coding agents](https://getconcord.ai/guides/handoff-work-between-ai-coding-agents): Move an unfinished task from one coding agent to another without losing what was tried, decided, changed, or left unresolved.
- [Coordinate coding agents across machines](https://getconcord.ai/guides/coordinate-coding-agents-across-machines): Keep local, cloud and remote coding agents aware of the same work even when they run in different terminals or environments.
- [Review and merge work from multiple AI coding agents](https://getconcord.ai/guides/review-merge-multiple-ai-coding-agents): Keep parallel agent changes understandable when several branches or worktrees converge into one codebase.

## Comparisons

- [Concord vs Conductor](https://getconcord.ai/compare/concord-vs-conductor): Conductor runs agents side by side. Concord makes them work together: who owns which file, what was decided, and what's ready to review.
- [Concord vs Traycer](https://getconcord.ai/compare/concord-vs-traycer): Traycer plans work and runs agents from its own app. Concord coordinates agents wherever they run, across harnesses, machines and teammates.
- [Concord vs Agent Relay](https://getconcord.ai/compare/concord-vs-agent-relay): Agent Relay gets agents talking. Concord gets them talking and stops them editing the same files while they do.
- [Concord vs Reason Network](https://getconcord.ai/compare/concord-vs-reason-network): Reason Network is a message line between agents. Concord is the shared work-state that keeps the conversation tied to tasks, files and evidence.
- [Concord vs Tutti](https://getconcord.ai/compare/concord-vs-tutti): Tutti asks your team to work inside its rooms. Concord coordinates agents where they already are: the terminal and the IDE.
- [Concord vs Amoeba](https://getconcord.ai/compare/concord-vs-amoeba): Amoeba wants to be your IDE. Concord works with every IDE and terminal agent your team already uses.
- [Concord vs OpenHands](https://getconcord.ai/compare/concord-vs-openhands): OpenHands runs agents. Concord makes agents, OpenHands' included, work as one team without colliding.
- [Concord vs Buzz](https://getconcord.ai/compare/concord-vs-buzz): Buzz is a chat room humans and agents share. Concord is built for the one thing coding agents get wrong together: the same repository.
- [Concord vs Alook](https://getconcord.ai/compare/concord-vs-alook): Alook puts your agents in a group chat. Concord gives them the shared work-state a group chat can't: who owns what, and what was decided.
- [Concord vs Mosaic](https://getconcord.ai/compare/concord-vs-mosaic): Mosaic records what your agents did. Concord coordinates what they're doing now, before two of them do the same thing.
- [Concord vs Linzumi](https://getconcord.ai/compare/concord-vs-linzumi): Linzumi lets a team steer agents from chat. Concord lets agents from every vendor coordinate with each other, with or without a human in the loop.

## Changelog

- [Concord 0.19](https://getconcord.ai/changelog/v0-19)
- [Concord 0.18](https://getconcord.ai/changelog/v0-18)
- [Concord 0.17](https://getconcord.ai/changelog/v0-17)
- [Concord 0.16](https://getconcord.ai/changelog/v0-16)
- [Concord 0.15](https://getconcord.ai/changelog/v0-15)
- [Concord 0.14: Global mode and project-scoped tasks](https://getconcord.ai/changelog/v0-14): Concord now works in repositories where concord setup was never run.
- [Concord 0.13: goose support and Windows fixes](https://getconcord.ai/changelog/v0-13): Concord adds an integration for goose, and the CLI can now spawn the concord executable on Windows.
- [Concord 0.12: Concord Cloud](https://getconcord.ai/changelog/v0-12): The first release that talks to Concord Cloud.
- [Concord 0.11: Quieter setup, faster status](https://getconcord.ai/changelog/v0-11): concord setup no longer edits your root .gitignore: the .concord/ directory ignores itself instead.
- [Concord 0.10: Reliable live messaging](https://getconcord.ai/changelog/v0-10): Live message delivery is reworked to be reliable across harnesses, and Concord now tracks when a receiving agent goes idle and recovers it.
- [Concord 0.9: Archive idle agents instead of dropping them](https://getconcord.ai/changelog/v0-9): Agents that haven't been seen for an hour are now archived rather than removed, so their record survives the end of the session.
- [Concord 0.8: Identity from the session](https://getconcord.ai/changelog/v0-8): An agent's identity now comes from its harness session instead of being minted fresh, so one session keeps one identity.
- [Concord 0.7: Messaging in plain claude and codex sessions](https://getconcord.ai/changelog/v0-7): Live messaging between agents now works in ordinary claude and codex sessions.
- [Concord 0.6: Five workflow tools](https://getconcord.ai/changelog/v0-6): The MCP surface is consolidated into the five workflow tools Concord uses today: start_work, inspect_work, update_work, transfer_work and finish_work.
- [Concord 0.5: Anonymous telemetry](https://getconcord.ai/changelog/v0-5): Concord starts collecting anonymous usage telemetry, including which CLI commands are run.
- [Concord 0.4: Presence and the live dashboard](https://getconcord.ai/changelog/v0-4): Concord learns who is online.
- [Concord 0.3: Task memory](https://getconcord.ai/changelog/v0-3): Tasks get their own memory: tools for recording what an agent learned while working, stored with the task instead of in the session.
- [Concord 0.2: Overlap checks at edit time](https://getconcord.ai/changelog/v0-2): Overlap detection gets sharper and moves closer to the edit.
- [Concord 0.1: First release](https://getconcord.ai/changelog/v0-1): The first published version: an MCP server with overlap detection on claim_work, handoffs and review-ready packets, backed by SQLite, plus the concord CLI with init, status, tasks, handoff, doctor and install.

## Topics

- [AI agent coordination](https://getconcord.ai/blog/topic/ai-agent-coordination) (1)
- [AI generated code technical debt](https://getconcord.ai/blog/topic/ai-generated-code-technical-debt) (1)
- [AI native engineering team](https://getconcord.ai/blog/topic/ai-native-engineering-team) (1)
- [AI pair programming](https://getconcord.ai/blog/topic/ai-pair-programming) (1)
- [Claude Code agent teams](https://getconcord.ai/blog/topic/claude-code-agent-teams) (1)
- [Claude Code and Codex together](https://getconcord.ai/blog/topic/claude-code-and-codex-together) (1)
- [best AI coding tools 2026](https://getconcord.ai/blog/topic/best-ai-coding-tools-2026) (1)
- [best MCP servers](https://getconcord.ai/blog/topic/best-mcp-servers) (1)
- [claude code features you didn't know](https://getconcord.ai/blog/topic/claude-code-features-you-didnt-know) (1)
- [claude code vs codex](https://getconcord.ai/blog/topic/claude-code-vs-codex) (1)
- [codex cli tips](https://getconcord.ai/blog/topic/codex-cli-tips) (1)
- [cursor rules](https://getconcord.ai/blog/topic/cursor-rules) (1)
- [cursor vs claude code](https://getconcord.ai/blog/topic/cursor-vs-claude-code) (1)
- [git worktrees multiple agents](https://getconcord.ai/blog/topic/git-worktrees-multiple-agents) (1)
- [how to become a 10x engineer](https://getconcord.ai/blog/topic/how-to-become-a-10x-engineer) (1)
- [how to make two AI agents work together](https://getconcord.ai/blog/topic/how-to-make-two-ai-agents-work-together) (1)
- [how to make your coding agent faster](https://getconcord.ai/blog/topic/how-to-make-your-coding-agent-faster) (1)
- [how to prompt a coding agent](https://getconcord.ai/blog/topic/how-to-prompt-a-coding-agent) (1)
- [how to review AI generated code](https://getconcord.ai/blog/topic/how-to-review-ai-generated-code) (1)
- [measuring developer productivity AI](https://getconcord.ai/blog/topic/measuring-developer-productivity-ai) (1)
- [multiple Claude Code agents](https://getconcord.ai/blog/topic/multiple-claude-code-agents) (1)
- [running multiple coding agents at once](https://getconcord.ai/blog/topic/running-multiple-coding-agents-at-once) (1)
- [share context between AI coding agents](https://getconcord.ai/blog/topic/share-context-between-ai-coding-agents) (1)
- [stop AI from hallucinating code](https://getconcord.ai/blog/topic/stop-ai-from-hallucinating-code) (1)
- [switching between AI coding tools](https://getconcord.ai/blog/topic/switching-between-ai-coding-tools) (1)
- [when not to use AI coding agents](https://getconcord.ai/blog/topic/when-not-to-use-ai-coding-agents) (1)

## Posts

- [Share context between AI coding agents](https://getconcord.ai/blog/share-context-between-ai-coding-agents): Share context between AI coding agents by persisting structured work-state, not just code. Use tools like `start_work` to register presence, `update_work` to log decisions, and `transfer_work` to hand off tasks cleanly.
- [Use Claude Code and Codex together on the same codebase](https://getconcord.ai/blog/use-claude-code-and-codex-together-on-the-same-codebase): You can run both Claude Code and Codex on one codebase by keeping separate workspaces, transferring ownership and assumptions between tools, and reviewing cross-tool changes before they reach review.
- [Agent teams in Claude Code what they are and when they help](https://getconcord.ai/blog/agent-teams-in-claude-code-what-they-are-and-when-they-help): Agent teams in Claude Code are groups of agents that coordinate on shared work, not just parallel sessions or nested subagents, and they need visible state to avoid duplicated or conflicting changes.
- [AI agent coordination keeps agents from stepping on each other](https://getconcord.ai/blog/ai-agent-coordination-keeps-agents-from-stepping-on-each-other): AI agent coordination is the shared workspace that tells each agent who owns which task, what is in scope, what decisions have been made and what others are waiting on, so independently running agents do not conflict.
- [Running multiple Claude Code agents on the same repo](https://getconcord.ai/blog/running-multiple-claude-code-agents-on-the-same-repo): You end up with a repeatable workflow that lets multiple Claude Code agents work on the same repository without duplicating effort or losing track of decisions.
- [How an AI native engineering team actually works](https://getconcord.ai/blog/how-an-ai-native-engineering-team-actually-works): An AI native engineering team scopes work in shared state, staffs reviews with captured decisions, and keeps coordination low by writing context into the work itself instead of external documents.
- [How to pair program with an AI coding agent](https://getconcord.ai/blog/how-to-pair-program-with-an-ai-coding-agent): Pairing with an AI agent requires tight loops and early signals of confusion to prevent wrong guesses from costing time.
- [Cursor rules that actually work for coding agents](https://getconcord.ai/blog/cursor-rules-that-actually-work-for-coding-agents): Write short, specific Cursor rules that target mistakes you correct repeatedly, and add checks that run outside the model to verify they are followed.
- [Codex CLI tips to keep your agent on task](https://getconcord.ai/blog/codex-cli-tips-to-keep-your-agent-on-task): Codex CLI tips help you scope tasks, keep runs short, read the plan before changes, and decide what the agent can do unsupervised.
- [When not to use AI coding agents](https://getconcord.ai/blog/when-not-to-use-ai-coding-agents): Use AI coding agents only when the cost of being subtly wrong, the time to review, or the external constraints do not outweigh the benefits.
- [Stop AI from hallucinating code in your repo](https://getconcord.ai/blog/stop-ai-from-hallucinating-code-in-your-repo): Make agents commit their assumptions and update them as the codebase changes so you catch hallucinations before they reach review.
- [Measuring developer productivity with AI](https://getconcord.ai/blog/measuring-developer-productivity-with-ai): Lines and PR counts became meaningless once both are cheap to produce, so better signals are review turnaround, rework rate, and overlap between agents on the same code.
- [Switching between AI coding tools without losing context](https://getconcord.ai/blog/switching-between-ai-coding-tools-without-losing-context): Switching between AI coding tools loses half-finished work unless you carry assumptions, rejected approaches, and changed files explicitly.
- [AI generated code technical debt](https://getconcord.ai/blog/ai-generated-code-technical-debt): Undocumented decisions are the main source of technical debt in AI generated code, not bad code itself.
- [How to prompt a coding agent for code changes](https://getconcord.ai/blog/how-to-prompt-a-coding-agent-for-code-changes): Write prompts that set constraints, name files, ask for a plan, and demand stated assumptions to get code changes you can review and trust.
- [Use git worktrees with multiple agents without merge hell](https://getconcord.ai/blog/use-git-worktrees-with-multiple-agents-without-merge-hell): Worktrees let you run two agents on the same repository without file-level conflicts, but they don’t stop agents from stepping on each other’s intent.
- [How to become a 10x engineer when code is cheap](https://getconcord.ai/blog/how-to-become-a-10x-engineer-when-code-is-cheap): Focus on reviewing and integrating generated work safely instead of typing speed. Use judgment about scope, taste in review, and documenting decisions so the next person is not guessing.
- [Best MCP servers for coding agents and what they do](https://getconcord.ai/blog/best-mcp-servers-for-coding-agents-and-what-they-do): MCP servers unlock repository access, issue tracking, browser automation, database queries, documentation search, and shared state between agents, each with different trade-offs in setup and trust.
- [How to make your coding agent faster](https://getconcord.ai/blog/how-to-make-your-coding-agent-faster): Narrow what the agent reads, give it the answer instead of letting it search, split a task so each step has a short context, and stop a run early when it has clearly misunderstood.
- [Cursor vs Claude Code for handing off work](https://getconcord.ai/blog/cursor-vs-claude-code-for-handing-off-work): Cursor keeps you in the editor with tight feedback for small tasks, while Claude Code takes longer tasks and explains its work in detail. Pick one based on how much you want to hand over at once.
- [How to review AI generated code](https://getconcord.ai/blog/how-to-review-ai-generated-code): Reconstruct intent from the stated goals, verify assumptions in the diff, treat unexplained changes as highest risk, and confirm tests cover the behaviour that changed.
- [Claude Code features you didn't know about](https://getconcord.ai/blog/claude-code-features-you-didnt-know-about): Five underused Claude Code features that let you plan before editing, share context across sessions, and end work deliberately instead of letting runs sprawl.
- [Running multiple coding agents at once without merge hell](https://getconcord.ai/blog/running-multiple-coding-agents-at-once-without-merge-hell): The ceiling is your review capacity, not compute. Split work so pieces do not touch, use separate checkouts, and watch for two agents with incompatible plans for the same interface.
- [How to make two AI agents work together](https://getconcord.ai/blog/how-to-make-two-ai-agents-work-together): Make the agents share a single source of truth by reading and writing to the same files, branches, or tools so they see each other’s intent and progress in real time.
- [Best AI coding tools for 2026 and what they’re for](https://getconcord.ai/blog/best-ai-coding-tools-for-2026-and-what-theyre-for): The best AI coding tools for 2026 are grouped by what you need them to do, not by a numbered ranking. Most teams end up with two or three tools rather than one.
- [Claude Code vs Codex for real work](https://getconcord.ai/blog/claude-code-vs-codex-for-real-work): Claude Code is better for unsupervised multi-file changes, while Codex is better for tasks that need steady steering. Choose based on your project’s coordination needs.