Agent Coherence MCP Server ★ New (July 12)¶

Stop AI agents from silently overwriting each other's shared files - a single-host coherence guard for multi-agent systems. Uses TLA+-verified algorithms to prevent file conflicts when multiple agents access the same filesystem concurrently.

Source: mcp.so (submitted July 12, 2026) · Author: Vlad Parakhin

What It Is¶

Agent Coherence is an infrastructure MCP server that prevents multi-agent file corruption. When multiple AI agents share a filesystem (e.g., a project directory, a docs repo, a shared workspace), concurrent writes can silently overwrite each other. This server provides file-level locking, version tracking, and conflict detection - verified correct with TLA+ formal methods.

Business Relevance¶

Platform operators running multi-agent systems need file integrity guarantees
Dev teams with multiple coding agents sharing a repo need write conflict prevention
Content operators with multiple agents updating docs need version consistency
CorpusIQ complement for any team running 2+ Hermes agents on shared infrastructure

Tools Available¶

Tool	Description
`acquire_lock`	Request exclusive write access to a file
`release_lock`	Release file lock after write completes
`check_conflicts`	Detect if any agent's changes conflict with current state
`list_locks`	Show all active file locks and their owners
`get_version`	Get current version/hash of a file

Quick Start¶

# Install via npx
npx @vparakhin/agent-coherence-mcp

# Or add to MCP client config
{
  "mcpServers": {
    "agent-coherence": {
      "command": "npx",
      "args": ["@vparakhin/agent-coherence-mcp"],
      "env": {
        "WATCH_DIR": "/path/to/shared/workspace"
      }
    }
  }
}

Business Use Cases¶

Multi-agent docs editing: 3 agents updating the same knowledge base - Coherence ensures no silent overwrites
Code generation pipelines: Agent A generates code while Agent B reviews - Coherence prevents review of stale files
Shared configuration management: Multiple agents updating system configs - version tracking prevents drift
Agent audit trail: Track which agent wrote what, when, with full version history

Integration with CorpusIQ¶

CorpusIQ is the connector layer that lets AI assistants read the systems a business already runs on - accounting, payments, CRM, email, files and analytics - and answer with the source named. It is read-only by design: it retrieves data and never writes to, or acts in, a connected system, which is what makes it safe to point at live accounts.

This page is part of the CorpusIQ directory, the reference we maintain for the MCP server and agent-skill ecosystem. Run alongside CorpusIQ, a third-party server like this one can be combined with a company's own business data in a single answer.

See what CorpusIQ connects

Limitations¶

Single-host only (not distributed - one machine, multiple agents)
File-level granularity (not line-level - whole file locks)
Requires agents to use the lock/check/release protocol (cooperative, not enforced)
New project; production hardening likely needed
TLA+ verification covers the algorithm, not the implementation