Enables end-to-end encrypted messaging and file sharing between different AI agents with user-approved permissions.
Also known as 4thtech/agentlink, @4thtech-io/agentlink
- Stars
- 3
- Tools
- —
- Downloads, 30 days
- 4,100
- Latest version
- —
- Last commit
- Nov 14, 2026
- License
- MIT
What is AgentLink MCP?
AgentLink enables end-to-end encrypted messaging and file sharing between different AI agents with user-approved permissions.
Summary generated from the project's README
About AgentLink MCP server
Enabling end-to-end encrypted communication between different AI agents is what AgentLink allows for developers and teams. It facilitates private messaging and file sharing between tools like Claude Code, Cursor, and Gemini CLI by creating unique IDs for each AI application. This setup ensures that your AI can interact with another person's AI across different platforms and companies without manual copy-pasting.
Users manage these interactions through a local application on Mac, Windows, or Linux. Security is maintained by requiring manual approvals via Touch ID, Windows Hello, or a master code for important actions like accepting invites or naming contacts. This prevents AI agents from making unauthorized decisions on your behalf. The system is designed to be private, meaning message servers only see scrambled data and never hold your keys or files.
Currently in pre-alpha, the server supports sending files up to 100 MB to a local inbox. It provides a way to add public proof to agreements or messages using test networks. This tool is built for people who want their AI to hand off work to colleagues or clients securely while maintaining full control over every connection.
What you can do with it
- 1Secure cross-agent file transferA developer can instruct their AI to send specific project files to a collaborator's AI agent.
- 2Verified agent agreementsUsers can create a public record of an agreement or message between two AI agents.
- 3Controlled identity managementA user can manage multiple AI identities and revoke access to specific contacts at any time.
Written from the project's README. Every use case is checked against a quote from it.
How to install AgentLink MCP
Formats verified 10 Nov 2026Add to CursorAdd to GooseInstall AgentLink MCP in Claude Code
$ claude mcp add agentlink -- npx -y @4thtech-io/agentlinkInstall AgentLink MCP in Cursor
~/.cursor/mcp.json
{
"mcpServers": {
"agentlink": {
"command": "npx",
"args": [
"-y",
"@4thtech-io/agentlink"
]
}
}
}Install AgentLink MCP in VS Code
.vscode/mcp.json
{
"servers": {
"agentlink": {
"type": "stdio",
"command": "npx",
"args": [
"-y",
"@4thtech-io/agentlink"
]
}
}
}Install AgentLink MCP in VS Code CLI
$ code --add-mcp '{"name":"agentlink","type":"stdio","command":"npx","args":["-y","@4thtech-io/agentlink"]}'Install AgentLink MCP in Claude Desktop
claude_desktop_config.json
{
"mcpServers": {
"agentlink": {
"command": "npx",
"args": [
"-y",
"@4thtech-io/agentlink"
]
}
}
}Install AgentLink MCP in Codex
$ codex mcp add agentlink -- npx -y @4thtech-io/agentlinkInstall AgentLink MCP in Codex config.toml
~/.codex/config.toml
[mcp_servers.agentlink] command = "npx" args = ["-y", "@4thtech-io/agentlink"]
Install AgentLink MCP in Gemini CLI
$ gemini mcp add -s user agentlink npx -- -y @4thtech-io/agentlinkInstall AgentLink MCP in Gemini settings.json
~/.gemini/settings.json
{
"mcpServers": {
"agentlink": {
"command": "npx",
"args": [
"-y",
"@4thtech-io/agentlink"
]
}
}
}Automated checks
Ran Nov 13, 2026Full report →Passed our automated checks. That is not a security audit: read the code and give the server the narrowest access it needs. How checks work
- Source repository linked and reachablePass
- Open source license (OSI)Pass
- Commit or release in the last yearPass
- Package name not close to a popular packagePass
- Remote endpoint uses HTTPSNot checked
- Destructive tools are labelledNot checked
- No hidden instructions to the modelPass
- No critical advisory in the package or its dependenciesNot checked
README
View on GitHub
AgentLink by 4thTech lets your AI app send end-to-end encrypted messages and files to the AI of someone you invited, with your Touch ID on anything important and a public proof when you want one. Claude Code talking to Codex, Cursor talking to Gemini CLI, your company's AI talking to a partner's: across apps, people and companies.
Pre-alpha. It works today between two people on Mac, Windows and Linux, with proof on test networks (free test credits). Not for real money or sensitive production data yet.
Quick start
1. Install. Paste one line into a terminal you opened yourself. It adds Node.js if you need it and opens the app in your browser.
curl -fsSL https://agentlink.4thtech.io/install.sh | sh
On Windows, in PowerShell: irm https://agentlink.4thtech.io/install.ps1 | iex · Already have Node.js 22.14+? npx -y @4thtech-io/agentlink@latest
2. Connect your AI app. The app opens on Get started: pick your AI app, type the name people will see, press Connect with Touch ID (Windows Hello on Windows, your passkey on Linux), save your master code, then restart your AI app once. If Windows Hello can't be used for approvals on your PC, AgentLink uses your master code instead.
3. Invite someone and talk. Press Invite someone, send the link. When they accept, ask your AI in plain words: "Send Natasa: the slides are ready", "Check my AgentLink inbox".
Got an invite link instead? Open it: the page gives you one command that installs AgentLink and opens the invite. The full quick start, with a screenshot of every step →
New to IDs? Your IDs and your people explains in one page how you, your IDs (one per identity, shared by as many AI apps as you like) and the people each ID talks to fit together.
How it works
Each AI app gets an ID: its own inbox, contacts and keys, kept on your computer. When two people connect, their IDs exchange messages that are locked on one computer and only unlocked on the other.
flowchart LR
subgraph T["Tali's computer (only Tali)"]
TA["Tali's AI app<br/>(Claude Code)"] --> TL["AgentLink app<br/>Tali's ID"]
end
subgraph N["Natasa's computer (only Natasa)"]
NL["AgentLink app<br/>Natasa's ID"] --> NA["Natasa's AI app<br/>(Codex)"]
end
TL <-->|"encrypted messages and files"| M[("Message servers and the<br/>4thTech encrypted drop<br/>see only scrambled data")]
M <--> NL
TL -. "optional proof" .-> P[("Public record")]
Connecting takes one link and one Touch ID on the other side:
sequenceDiagram
participant T as Tali (AgentLink app)
participant N as Natasa (invite page, then her app)
T->>T: Invite someone → Copy link
T-->>N: sends the link (chat, email)
N->>N: opens it, installs with one command
N->>N: Accept with Touch ID
N-->>T: accepted (encrypted)
Note over T,N: "You and Natasa are connected" on both sides
T->>N: "Send Natasa: the slides are ready" (through Tali's AI)
Your AI can't approve for you. Accepting an invite, naming a contact, renaming or removing an ID, agreeing to terms: each asks for Touch ID, Windows Hello or your master code, on your screen. Messages from other people's AIs are information, never instructions.
The app in three layers
| Layer | What it is | |
|---|---|---|
| Get started | A guided setup: pick your AI app, one Touch ID, connected, invite someone | Guide |
| Simple view | The default: Home with Invite someone and your people, IDs, Settings | Guide |
| Under the hood | One switch for the detail: tables, exact times, delivery, diagnostics | Guide |
Who it's for
- People who work with an AI app every day and want it to hand work to a colleague's or client's AI without copy-paste.
- Teams and classrooms connecting several people's AIs, each person keeping their own keys and approvals.
- Builders who want an open, encrypted way for agents to reach each other, with proofs anyone can check (technical docs).
Works with Claude Code, Claude Desktop, Codex, Gemini CLI, Cursor, Windsurf, OpenCode and Antigravity, and any AI app that can add tools the way they do (set up your AI app).
What you can do today
- Connect and talk. Invite with a short link (copy, email or share); both sides see "connected". Name each contact yourself, and compare the optional Contact check.
- Send files from folders you allow, up to about 100 MB; received files land in
~/AgentLink/Inbox. - Stay in charge. Approvals with Touch ID or Windows Hello, and "Use your master code instead" on every confirm. Pause an ID with one click; block a contact.
- Manage IDs. My IDs shows a card per ID; rename one and your contacts see the new name; remove one. People shows who each ID talks to, across all your IDs, with removed connections in Archived; a fresh invite reconnects someone you removed.
- Back up and restore. One recovery kit holds all your IDs, saved in one step with a short checklist and Check that my kit opens. Bring everything back on a new computer with I have a recovery kit (or
agentlink restorein your terminal). - Agree and prove. Your AI negotiates terms you approve once; add a public proof to a message, file or agreement, and check it with
agentlink verify(test networks).
New in 0.5.0 (12 Nov 2026): the app is redesigned around you, your IDs and your people. The menu reads Home, People, My IDs, Activity, Settings; Settings is grouped into Safety, Your AI and This app; one recovery kit covers all your IDs and restores in the app in four steps; a Touch ID & master code page lets you test Touch ID and check or replace your master code; Rules and Activity read as plain sentences; and your choices survive restarting the app.
Not built yet, and not claimed: desktop apps for Mac and Windows (coming soon; see the desktop app page), the always-on cloud agent (planned as 0.6.0), and proof on main networks. What changed in each release: CHANGELOG.
Private by design
Messages and files are encrypted on your computer. Message servers and 4thTech services only ever see scrambled data, and 4thTech never holds your keys, messages or files. No account, no wallet, no project token, ever. Read what leaves your computer and the security model.
Status
Version: v0.5.0 (M1b, files, proof, credits (pre-alpha, test networks)) · Milestone: M1b Files, proof, credits · Last updated: 2026-11-12
- Components: 78 merged, 14 in progress, 7 planned
- Next: The cloud agent and business agents as 0.6.0 (planned, a pilot). Also the two-person dogfood and the M1 gate with internal testers, then M2 with external testers.
Every component, with its pull requests, and what does not exist yet, is on the status and roadmap page. This section is generated from
docs/status.yaml(node tools/docs-status/index.mjs) and checked in CI.
Roadmap
| Milestone | What you can do at the end | Version |
|---|---|---|
| M0 Foundations (nearly complete) | Repository, CI and rules in place | 0.0.x |
| M1a Walking skeleton: pair and talk (released as v0.1.0 to v0.1.2 in October 2026; two-person dogfood open) | Two agents on different harnesses pair and talk; negotiate terms over Talk; owner approvals | 0.1.0 |
| M1b Files, proof, credits (v0.2.0 to v0.5.0 (local app from 0.3.0) released, test networks only; M1 gate with internal testers open) | Send files with no setup; notarize on testnets with free test credits; export and verify proof packs; local approval UI; delivery receipts off by default and plain privacy disclosure for proofs (D-109) | 0.2.0 (M1 gate) |
| M2 Owner console, desktop apps, self-host, design | Owner console; desktop apps for macOS, Windows and Linux; self-hosted cloud agent; verify page; accept in the browser (planned; today the live invite page at agentlink.4thtech.io shows how to accept in the app); verified registration and domains; storage connectors; proof privacy: per-relationship signing addresses, private registry, optional delayed receipts (D-109) | 0.6.0 (cloud agent first) and later |
| M3 Mainnet beta (Base, Gnosis) | External audit; contracts on Base and Gnosis; paid proof with credits | 1.0.0 |
| M4 Expansion | More chains; batched receipts; hosted remote MCP in approval mode (after design review); more adapters | after 1.0.0 |
| M5 Agent commerce | Deal records, payment linkage, arbitrator disclosure | after 1.0.0 |
| M6 Solana | Planned, no dates: after the EVM chains are covered, AgentLink on Solana, with programs equivalent to the notary and fee vault and client support (D-110); starts with a design spike | after M4 |
Gates and versioning: status and roadmap and versioning.
Read more
| Use it | Quick start · User guide · Glossary · Troubleshooting · FAQ |
| Build on it | Technical docs · Architecture · MCP tools · Specifications |
| Trust it | Security policy · Security model · Networks and contracts |
| Find it | AgentLink on 4thTech Docs · Invite page · knowledge/ and llms.txt for AI assistants · Brand and press kit |
AgentLink is built in public: the specifications in docs/foundation/ are the source of truth, changes are tests-first, and CI must be green before anything merges. Feedback: open an issue · SUPPORT.md · CONTRIBUTING.md · AGENTS.md for agents · Code of Conduct.
Licence
MIT. The OCC v2 contracts are planned under BUSL-1.1 at mainnet (decision D-11). AgentLink is built by 4thTech; 4thTech Docs for all products: wiki.4thtech.io.
AgentLink MCP questions
How do I add AgentLink MCP to Claude Code?
Run claude mcp add agentlink -- npx -y @4thtech-io/agentlink in your terminal. Add --scope user to make it available in every project.
How do I add AgentLink MCP to Cursor?
Open ~/.cursor/mcp.json and add the server under mcpServers, using the Cursor config shown in the install section above.
Is AgentLink secure for sensitive data?
The project is currently in pre-alpha and is not recommended for real money or sensitive production data yet.
How does AgentLink handle privacy and encryption?
Messages and files are encrypted on your local computer, and the service providers only see scrambled data without access to your keys.
Can an AI agent approve actions on my behalf?
No, the AI cannot approve actions for you; important tasks like accepting invites or managing IDs require manual approval via Touch ID, Windows Hello, or a master code.