Senrelay
中EN
GitHub (opens in a new window)
DECISION GUIDE / MULTI-AGENT CONTEXT & MCP

5 Ways to Share Context, Memory & MCP Across AI Coding Agents (2026)

No single tool fits every workflow when switching between Claude Code, Cursor, Codex, Antigravity, OMP, and OpenCode: cross-platform teams on Windows or Linux should start with repo-committed CLAUDE.md + AGENTS.md files ($0); multi-machine cloud sync fits hosted Memory MCP servers (free tier or paid plans); and single-machine macOS 14.0+ (Apple Silicon) developers switching multiple agents mid-task can pair repo files with Senrelay ($0, MIT open source, no signup).

Prices & technical specs verified: Checked on 2026-11-11 By Qingsong Cui (George O'Nair) Fairness rule: Senrelay not ranked #1 · real downsides listed for every option
✦ 30-Second Verdict (TL;DR: Pick by Your Actual Setup)
  • On Windows, Linux, or Intel Mac—or sharing team conventions in Git → Use Option 1: Repo-committed CLAUDE.md + AGENTS.md ($0). Skip Senrelay (it requires Apple Silicon macOS).
  • Syncing memory across multiple machines (e.g. Linux desktop + MacBook) → Use Option 2: Cloud or Docker Memory MCP servers (e.g. Mem0 / OpenMemory).
  • Working on a single Apple Silicon Mac (macOS 14.0+) across Claude Code, Cursor, Codex, Antigravity, OMP, or OpenCode with strict local privacy → Combine Option 1 + Option 3 (Senrelay, $0 MIT open source).
  • Staying inside one IDE (e.g. Cursor only) and just switching underlying LLMs → Use Option 5: Single IDE + Multi-Model Router (e.g. OpenRouter)—you do not need cross-agent relay software.

1. 30-Second Interactive Stack Picker Interactive Quiz

Select your operating system, privacy boundary, and daily agent workflow to see which architecture fits—including when you should not install Senrelay:

① Primary Operating System
② Data & Sync Boundary
③ Daily Coding Workflow
Recommended Stack: Option 1 (AGENTS.md Static Rules) + Option 3 (Senrelay Local Workbench, $0)

For single-Mac Apple Silicon developers switching across multiple agent harnesses with strict local privacy: keep shared repo rules in AGENTS.md, and run Senrelay on 127.0.0.1:8420 for /contextcore session relay, source-linked 💡 auto-memory, and safe ~/.claude.json + ~/.codex/config.toml sync.

2. Side-by-Side Technical Matrix Checked on 2026-11-11

Comparison of 5 Cross-Agent Context, Memory & MCP Sync Approaches (Checked on 2026-11-11)
Approach OS Support Official Pricing (Plain Text) Mid-Task Session Relay Source-Linked Auto-Memory Multi-Host MCP & Skills Sync Data Residency
1. Repo Rule Files (CLAUDE.md / AGENTS.md) All OSes (Win / Mac / Linux) $0 (built-in) No (static rules only; misses live chat state) No (manual Markdown edits) No (does not manage global host MCP configs) Local Git repository
2. Cloud / Docker Memory MCP (Mem0 / OpenMemory) All OSes (any MCP client) Self-hosted $0 / Cloud free tier & paid plans Partial (retrieves memory facts, no numbered transcript handoff pack) Semantic extraction (typically without local turn-level transcript links) No (must configure MCP manually per host) Cloud API or local Docker stack
3. Senrelay (ContextCore Local macOS Workbench) macOS 14.0+ Apple Silicon (arm64) only $0 (100% Free & Open Source, MIT License, no signup) Native (/contextcore lists recent sessions across 6 hosts; brief-only or continue) Native (companion capture + 💡 receipt + verbatim provenance + conflict inbox) Native (6 core + 9 extended hosts with diff guard & rollback snapshots) 100% local SQLite (0600) + macOS Keychain
4. Dotfiles & Sync Scripts (GNU Stow / Custom CLI) macOS / Linux (custom scripts on Windows) $0 (DIY open source) No No Fragile (JSON vs TOML vs JSON5 vs YAML schema mismatch breaks symlinks) Local Dotfiles repo (risk of committing secrets)
5. Single Multi-Model IDE / Router (Cursor / OpenRouter) All OSes (depends on IDE) IDE plan ($0–$20+/mo) + token usage Same-window only (cannot relay across separate agent tools) Varies by single IDE Single host only Local + chosen model router

3. Honest Breakdown of All 5 Options Senrelay Ranked #3, Not #1

To keep this guide genuinely useful, we rank the universal zero-dependency standard first, place Senrelay at #3, and document concrete downsides for every single option—including our own.

Option 1: Repo-Committed Instruction Files (CLAUDE.md + AGENTS.md + .cursor/rules)

Pricing: $0 · Universal Cross-Platform Standard

Check project architecture rules, test commands, and guardrails straight into Markdown files at the root of your repository. Claude Code reads CLAUDE.md natively while Codex, Cursor, OpenCode, and most modern agents read AGENTS.md. Almost every project should start here.

✦ Best For

Cross-platform engineering teams on Windows, Linux, and macOS who want static repository rules versioned alongside pull requests.

⚠ Real Downside

Static files do not capture live session state. If you spend 40 minutes debugging a race condition in Claude Code and switch to Cursor or Codex when hitting a rate limit or UI task, AGENTS.md has none of your recent conversation turns, ruled-out hypotheses, or global MCP server definitions.

Option 2: Cloud or Docker-Hosted Memory MCP Servers (e.g. Mem0 MCP / OpenMemory)

Pricing: Self-hosted $0 / Hosted free tier + paid plans (Checked on 2026-11-11)

Connects an external semantic memory server over MCP so multiple clients can store and query extracted facts through a shared cloud API or local Docker stack (vector DB + API service).

✦ Best For

Developers moving constantly between a Linux workstation, Windows PC, and laptop—or teams building cloud-hosted agent applications.

⚠ Real Downside

Hosted plans send extracted project context to external servers, while self-hosting requires maintaining Docker containers. Generic memory MCPs also do not index native local agent transcript files for numbered session handoff or manage host provider/Skills config files.

Option 3: Senrelay (ContextCore Local macOS Multi-Agent Workbench)

Pricing: $0 (100% Free MIT Open Source · No Signup · Checked on 2026-11-11)

A native macOS app and loopback service (127.0.0.1:8420) uniting local transcript discovery, /contextcore 6-point cross-agent conversation relay, companion source-linked auto-memory (with a brief 💡 receipt), and safe multi-host MCP, Skills, and provider configuration sync across Claude Code, Codex, Cursor, Antigravity, OMP, and OpenCode (see the 6+9 host matrix and setup guide).

✦ Best For

Developers working on a single macOS 14.0+ (Apple Silicon) machine who bounce between 2+ AI coding agents daily, require 100% local SQLite + Keychain privacy, and want one place to sync MCP servers, Skills, and provider presets.

⚠ Real Downside (Read Before Installing)

(1) macOS Apple Silicon only: No Windows, Linux, or Intel (x86_64) Mac support; (2) No cloud sync: Single-Mac loopback design—moving Macs requires contextcore backup; (3) Extended host scope: 9 extended hosts support config management only, not session relay; (4) First-launch step: Public Beta uses ad-hoc signing without Apple notarization, requiring a one-time xattr -cr /Applications/Senrelay.app.

Option 4: Dotfiles Symlinks & Custom Config Scripts (GNU Stow / jq-toml Scripts)

Pricing: $0 · DIY Terminal Workflow

Uses a personal Dotfiles Git repo with GNU Stow or custom shell scripts to version global prompt files, Skills folders, and MCP definitions across ~/.claude/, ~/.codex/, and ~/.cursor/.

✦ Best For

Terminal minimalists on Linux or macOS who only care about static config files and never need mid-task conversation relay or searchable local memory.

⚠ Real Downside

Incompatible formats—Claude Code uses JSON (~/.claude.json), Codex uses TOML (~/.codex/config.toml), OpenClaw/Pi use JSON5, and Hermes uses YAML—so raw symlinks fail. Worse, agents frequently rewrite their own config files with runtime state, breaking symlinks or risking accidental API key commits.

Option 5: Staying Inside One Multi-Model IDE or API Router (e.g. Cursor / OpenCode + OpenRouter)

Pricing: IDE Subscription + Pay-Per-Token API (Checked on 2026-11-11)

Keep all coding inside a single editor or CLI harness and swap underlying models (Claude, GPT, Gemini, DeepSeek) via built-in model selectors or an OpenRouter key instead of switching tools.

✦ Best For

Developers satisfied with one single UI who only want model diversity rather than combining specialized agent harnesses.

⚠ Real Downside

Modern coding agents differ dramatically in harness capabilities—Claude Code's autonomous CLI loop, Cursor's multi-file tab/diff UX, Codex's sandboxed execution, and Antigravity's browser/desktop workflows. Locking into one client means losing context the moment a subscription quota resets or another harness fits a task better.

4. How We Tested & Verified Show the Method

Every specification on this page was verified directly against local macOS 14+ (arm64) installations ():

  • Native Host Config Inspection: Verified schema serialization, external-edit conflict detection, and snapshot rollback across ~/.claude.json (JSON), ~/.codex/config.toml (TOML), Cursor MCP configs, and .gemini/settings.json (see the homepage 6+9 host matrix).
  • Relay Execution Safety Check: Verified on 127.0.0.1:8420 that number-only selection enforces must_not_execute=true (summarizing context without modifying files), while attaching an explicit task executes immediately using the 6-point relay pack.
  • Zero Inflated Claims: Pricing was checked against public documentation, and all 4 Senrelay limitations (Apple Silicon requirement, single-Mac local storage, config-only extended hosts, and Beta xattr -cr first launch) match our automated repository test suite.

5. Decision FAQ Conclusion-First Answers

If I already maintain CLAUDE.md and AGENTS.md, do I still need Senrelay?

No if you only need static repo conventions; yes if you switch agents mid-task or manage MCP servers across multiple hosts—they solve different layers and work best together. AGENTS.md stores timeless repo rules in Git, whereas Senrelay relays your live 30-turn debugging context, captures source-linked memories (with a 💡 receipt), syncs global MCP/Skills configs, and can export pinned rules back into AGENTS.md.

What is the fastest way to switch from Claude Code to Cursor or Codex without losing context?

On Apple Silicon macOS, type /contextcore (or "Relay") in your target agent with Senrelay running to pick your recent session by number. On Windows or Linux, ask your current agent to write a concise HANDOFF.md summarizing goals, completed diffs, ruled-out approaches, and next steps before opening the next tool.

When should I choose Mem0 or cloud memory MCP over Senrelay?

Choose Mem0 or a cloud/Docker memory MCP whenever you work across Windows, Linux, or multiple computers, or when building cloud-hosted multi-user agent apps. Senrelay is purpose-built for single-Mac local developer privacy and intentionally omits cloud storage.

How can I safely sync MCP servers between Claude Code and Codex without breaking configs?

Translate entries natively between JSON (~/.claude.json) and TOML (~/.codex/config.toml) with pre-write snapshots—never symlink one file to the other. Senrelay's Agent Management converts formats automatically, isolates API keys in macOS Keychain, and opens a side-by-side Diff prompt whenever a host config was edited externally (read security boundaries).

Is Senrelay genuinely $0 forever, and how do I install it on macOS?

Yes—Senrelay is 100% free ($0) under the MIT License with zero account signup or paid tiers. Install on macOS 14.0+ (Apple Silicon) via brew install --cask qingsongcui/tap/senrelay and run xattr -cr /Applications/Senrelay.app once on first launch (full setup guide; never run the unrelated PyPI pip install contextcore package).

Author & Editorial Disclosure

Written and maintained by Qingsong Cui (George O'Nair) (GitHub @qingsongcui · X @George_ONair), creator of the open-source Senrelay (ContextCore) project. Because we build Option 3, we deliberately do not rank ourselves #1 and state upfront whenever another workflow is a better fit.

Try Locally or Inspect Source ($0 MIT Open Source)

On macOS 14.0+ (Apple Silicon), install via brew install --cask qingsongcui/tap/senrelay, or inspect the Local Setup Guide and Homepage Official Fact Sheet.