M
MCP Directory CURATED DIRECTORY
911 MCP servers indexed
中文
← Back to directory
A

AGNT

Community
Not a standard server
The local-first agent operating system for building, running, and improving AI agents, workflows, and autonomous goals.

This is not a typical connect-and-use MCP server; read the source repository before using it.

How to install ↓ Compare GitHub source repository ↗
Category
Dev Tools #412 of 438
Stars
★ 553 Very popular
Transport
stdio (local process)
Runtime
Node.js 18+ · Docker
Credentials
Optional API key
License
Other / unspecified
Last commit
Tools
9
30FMRS · D

AGNT is a feature-rich local-first agent operating system offering agents, workflows, goals, and plugin system. It emphasizes repeatability and self-improvement, suitable for users needing a persistent AI work platform. With local-first design, data control is strong, but technical knowledge is required for setup and configuration.

Strongest · Maintenance 8/20 Weakest · Reliability 4/20

Reliability
4/20
Security and permissions
7/20
Maintenance
8/20
Documentation
7/20
Setup experience
4/20
Why each score
Reliability 4/20
Only the README was available; no MCP server manifest, source tree, tests, or CI logs could be inspected. The README claims a local Express backend, SSE, workflows, and MCP support, but nothing demonstrates an MCP initialization handshake or tool-list return at the repository root. Deducted for missing executable evidence and absent MCP protocol details.
Security and permissions 7/20
No malware or real secrets were found; the README describes local-first storage and API keys being sent only to configured providers. However, the local API surface is broad, and the .env.example implies default JWT/SESSION/ENCRYPTION_KEY values must be changed in production; no least-privilege model, no confirmation flow for destructive actions (delete/pay/deploy/remote commands), and no MCP permission boundary are documented. Deducted for missing permission/confirmation evidence and default-secret risk.
Maintenance 8/20
The repository is not archived, the README shows version 0.6.5, tag-driven release scripts, and CI/CD documentation; however, no commit history, dates, issue-response evidence, dependency-update policy, or security-response channel were provided. The license is declared as Custom License and GitHub reports NOASSERTION, weakening verifiable governance. Deducted for missing maintenance evidence and ambiguous license.
Documentation 7/20
The README is comprehensive for the AGNT application itself (install modes, architecture, providers, plugins, API families). But MCP-server specifics are absent: no server manifest, no client configuration example, no MCP tool schemas/limits, and no MCP troubleshooting. Deducted for missing MCP-layer documentation and unverified linked docs.
Setup experience 4/20
Install steps for the application are given (Node 18+, npm install, frontend install, npm start, port 3333), but there is no MCP client setup (e.g., JSON configuration), no one-command MCP launch, and no proof of a successful MCP connection. Production requires building the frontend and configuring secrets. Deducted for manual multi-step setup and no MCP client examples.

Static review · not runListed 2026-08-07

Read the FMRS scoring method →

Fit and risk

What it can accessReads local filesWrites / deletes local filesRuns commands or codeUses the networkControls a browserConnects to a database

Best for

  • Individual users running AGNT on a personal machine.
  • Small teams of 2-10 people sharing one backend.
  • Developers wanting a local agent backend they can inspect and extend.
  • Automation builders needing visual workflows plus AI judgment.
  • Power users wanting durable agents instead of throwaway chats.

Not for

  • Public multi-tenant SaaS.
  • Hosting unrelated organizations on one shared instance.
  • Large enterprises with 50+ concurrent users or zero-trust isolation requirements.
  • Users who only need a hosted chatbot or prompt playground.

Required permissions

  • Access local filesystem to read code and files.
  • Execute custom code (JavaScript/Python).
  • Call external APIs including AI providers.
  • Send and receive messages (Telegram, Discord, etc.).
  • Access network to connect to services.

Risks and side effects

  • Local API on port 3333 might be exposed to network, risking unauthorized access.
  • Ensure production secrets are changed for secure deployments.
  • Plugins can contain arbitrary code, so only install from trusted sources.
  • Browser automation may consume significant resources.

Setup

Before you start

Runtime：Node.js 18+ · Docker

JWT_SECRET optionalsecret JWT signing secret, generate with openssl rand -base64 32; must be changed in production.
SESSION_SECRET optionalsecret Session signing secret; must be changed in production.
ENCRYPTION_KEY optionalsecret Encryption key for local data; must be changed in production.
Other optional settings (4)
AGNT_HOME optional Overrides the default data directory (defaults to ~/.agnt); set an absolute path for Docker snap installs.
NODE_ENV optional Runtime environment flag; set to production for deployments.
AGNT_LITE_MODE optional Set to true to run Lite mode without browser automation.
PORT optional Local backend listen port, defaults to 3333.

Clone the repository: git clone https://github.com/agnt-gg/agnt.git; change directory: cd agnt; install dependencies: npm install; install frontend dependencies: cd frontend && npm install && cd ..; start: npm start. The backend runs at http://localhost:3333.

Check that it works

Open the AGNT dashboard at http://localhost:3333 and confirm the agents/workflows modules load; in your client's tool list you should see agent, workflow, goal and the other recorded tools.

Troubleshooting

  1. Ensure Node.js version is >=18.
  2. Check if port 3333 is occupied.
  3. Verify environment variables like JWT_SECRET are set correctly.
  4. Check log files (e.g., <data-dir>/_logs/) for errors.
  5. For Docker, use --pull always to fetch latest image.

Things to try

Once connected, you can ask your AI assistant things like:

  • Create a new agent that organizes my GitHub repository information
  • Start a goal to fetch the news every morning and generate a summary
  • Show me the run status of all my workflows
  • Turn this execution trace into a reusable skill

Tools 9

agent writes
Provides creation, management, and chat with persistent AI workers.
workflow writes
Provides visual DAG automation workflows for creation, start, stop, and status.
goal writes
Manages autonomous objectives including planning, execution, evaluation, and rollback.
skill writes
Manages reusable skills and learning from execution traces.
plugin writes
Installs, uninstalls, reloads plugins and manages marketplace assets.
widget writes
Creates and manages custom dashboard widgets.
experiment writes
Runs evaluations and benchmark tests.
insight writes
Extracts and views learned insights and memory from runs.
Show 1 more tools
mcp_server writes
Manages MCP servers and capability discovery.

Use cases

Build repeatable automation workflows with AI judgment.
Run long-term autonomous goals like a product launch.
Delegate tasks across multiple agents, tools, and APIs.
Run AI work locally or self-hosted with durable, inspectable traces.
Create reusable skills from execution traces.

Supported clients

Claude Desktop

Listed from the project's documentation, not tested by this site.

Overview

AGNT is a local-first agent operating system, providing a desktop app and self-hostable backend for building, running, evaluating, and evolving AI work. It brings agents, workflows, goals, tools, memory, plugins, evals, traces, messaging, and APIs into one local workspace. AGNT supports 15+ AI providers, 60+ built-in tools, 25+ plugin templates, and is MCP-compatible. Its core execution modes are Agents (for judgment), Workflows (for repeatability), and Goals (for long-running objectives). AGNT is local-first, storing data in SQLite and filesystem, with inspectable traces, costs, and decisions.

Similar servers

Context7 80 · B

Upstash's official server providing up-to-date third-party library docs for AI coding assistants

★ 62.9k · Tools 2 Compare with this →
Compare with other MCP servers

Source revision 52f1a8644bf8 Data synced 2026-10-12 Read the FMRS scoring method