File-claim coordination for parallel AI coding agents: claim files before edits, no stomping.
Copy the AI prompt to install this server into Claude Code, Cursor, or another agent — or use 1-click editor setup below.
💡 Paste the JSON block into your client's configuration file under mcpServers, then restart the application.
Languages: English · 日本語 · 简体中文 · Русский · Español
Use Agent Claim MCP when multiple coding agents share one worktree and you need a tiny local coordination primitive before edits, not a full orchestration framework. It gives agents one job: claim paths, detect collisions, and release ownership so parallel work stops stomping the same files.
Agent Claim MCP is a local-first MCP server for Claude Code, Cursor, Cline, and other MCP clients that need file ownership coordination without queues, planners, or custom AGENTS.md conventions. The current release surface centers on three bounded actions only: claim normalized paths, inspect who owns them, and release by path or claim id with explicit conflict reporting across separate sessions.
Release status:
@vk0/agent-claim-mcp@1.0.0is live on npm, so thenpx -y @vk0/agent-claim-mcpinstall path is now the truthful default for external users. Phase 5 external proof may be satisfied by either Official MCP Registry validation or a live Smithery listing, but the current proof state remains pending until one of those public artifacts is explicitly verified. Official MCP Registry validation is still pending, so do not describe the package as registry-accepted or marketplace-listed until a real registry or Smithery proof URL is cited.Milestone truth: the product wedge stays narrow, 3 tools plus a local JSON ledger for shared-worktree collision prevention, not a broader orchestration platform.
Current next manual step: complete the pending human rerun and verification path for Official MCP Registry proof, then record the verdict from that run instead of assuming acceptance from npm availability.
Current truthful state:
1.0.0 is live25282612113 did not complete the registry validation pathNPM_TOKEN secret-fix step, but the registry proof still needs a human rerun plus verification before it can be treated as acceptedExact rerun command:
After the rerun, verify in this order:
npm run preflight:registrydocs/official-registry-validation-runbook.mdOperator shortcut: use the runbook's Quick operator path for the exact rerun-to-verification order, and treat npm-live status as separate from registry-proof status.
Choose this server when your workflow already has task routing, but still needs a simple lock-like primitive at edit time:
The canonical external install path is now the published npm package:
Status note: npm install is live today, but Official MCP Registry validation and Smithery/other marketplace proof are still not established, so treat the package as npm-available first rather than marketplace-verified.
If you are developing from a local checkout instead of the published package, you can still point your MCP client at the built dist/server.js directly.
That second example is for local development only. On Windows, point your MCP client at the built dist/server.js path on your own machine instead of copying a POSIX path literally.
Build once before using the local path:
Add the stdio server to your Claude Code MCP config:
Claude Desktop uses ~/Library/Application Support/Claude/claude_desktop_config.json on macOS and %APPDATA%\Claude\claude_desktop_config.json on Windows:
After saving claude_desktop_config.json, fully restart Claude Desktop so it reloads the MCP server configuration.
This server is intentionally narrow and local-first. Install it when your agents share one filesystem view, not when you need distributed coordination.
cwd for claims to line up.For a real multi-agent run before publishing, see DOGFOOD.md and the proof artifacts in docs/dogfood-report.md.
claim_filesCreate or refresh a local file-claim entry so parallel agents can see ownership before editing and avoid stomping the same worktree paths during active tasks. Conflict responses are explicit, so overlapping claims across separate sessions do not stay implicit or in-memory only.
Input
Output
If any requested path is already claimed by another active owner, the tool returns ok: false and writes nothing.
release_claimRemove an existing claim owned by the current agent so finished, paused, or reassigned work stops blocking other agents from safely editing the same paths.
Input
Output
Only the current owner can release an active claim, whether you target it by claimId or by normalized paths, so cleanup works after real multi-agent runs and path-shape variants resolve consistently.
whose_claimRead the local ledger and explain whether a file path is free or currently claimed, including owner, task, note, and expiry metadata for safe coordination across separate sessions sharing the same worktree.
Input
Output
foo.tsclaim_files returns ok: falseNo reviews yet — be the first to share how this listing worked for you.
Showcase your server listing on GitHub or your project documentation. Embed this dynamic SVG badge to highlight official listing status and live engagement.
[](https://allmcps.com/mcp/agent-claim-mcp)<a href="https://allmcps.com/mcp/agent-claim-mcp"><img src="https://allmcps.com/api/badge/agent-claim-mcp?style=directory" alt="Agent Claim MCP on AllMCPs" /></a>