Atlas / MCP servers / anis-marrouchi / agentix-cli

agentix-cliBLOCK

Name: agentix-cli
Author: anis-marrouchi

mcp/anis-marrouchi/agentx

MCP server by anis-marrouchi · agent · ai agents · anthropic ai sdk · Audited 2026-10-11

Self-hosted AI agents for small teams. Route Telegram, WhatsApp, GitHub, GitLab, crons, webhooks and mesh tasks to agents on Claude Code, Codex, OpenCode or any LLM API. Web dashboard + CLI. MIT.

Verdict

BLOCK

Grade

Trust score

32 /100

Exposed tools

121 89r · 30w · 2d

Transport

—

License

MIT

Stars

Overview

From the repository's own README, as read at the audited commit. Badges and raw HTML are left out.

Put an AI teammate on the tools your team already uses.

[](https://www.npmjs.com/package/agentix-cli) [](https://github.com/anis-marrouchi/agentx/actions/workflows/ci.yml) [](LICENSE) [](package.json)

Watch it · See the demo · Install · Read the guide

AgentX routes messages and scheduled jobs to AI agents you host. Connect Telegram, WhatsApp, GitLab, GitHub, or a webhook. Give each agent a clear job. Watch the result in a browser dashboard.

Watch it

Short videos show AgentX running, one task each. Click the picture to open the Learn AgentX playlist on YouTube.

[](https://www.youtube.com/playlist?list=PLWyRWfXkYvlw)

Read from source at commit b5d4850b6beeOBSERVED · 2026-10-11

Connect

Built from this server's own package name, version and transport as found in its source — not copied from anyone's documentation, so it cannot drift against a page we do not control. Replace the environment placeholders with a token scoped to the least it needs.

claude-code

claude mcp add agentx-plugin-echo-channel --env ACME_AGENT_TOOLS_SECRET=${ACME_AGENT_TOOLS_SECRET} --env AGENTX_CHAT_SECRET=${AGENTX_CHAT_SECRET} --env AGENTX_MESH_AUTH=${AGENTX_MESH_AUTH} --env AGENTX_TOKEN=${AGENTX_TOKEN} -- npx -y [email protected]

claude-desktop

{
  "mcpServers": {
    "agentx-plugin-echo-channel": {
      "command": "npx",
      "args": [
        "-y",
        "[email protected]"
      ],
      "env": {
        "ACME_AGENT_TOOLS_SECRET": "${ACME_AGENT_TOOLS_SECRET}",
        "AGENTX_CHAT_SECRET": "${AGENTX_CHAT_SECRET}",
        "AGENTX_MESH_AUTH": "${AGENTX_MESH_AUTH}",
        "AGENTX_TOKEN": "${AGENTX_TOKEN}"
      }
    }
  }
}

Exposed tools (121)

89 read · 30 write · 2 destructive. Blast radius: 2 tools can delete or overwrite — an agent that can be talked into calling a tool can be talked into calling this one.

Tool	Risk	Description
`AgentX`	read	Talk to your agents and watch your fleet.
`Bad-Name`	read	x
`Main`	read	The first computer
`add_deliverable`	write	Record a shipped artefact against a task (repo URL, preview URL, design file, document, etc.). type: repo\|preview\|design\|artifact\|document\|screenshot\|figma\|spreadsheet\|presentation\|video\|audio\|image\|other.
`admin`	read	Administration
`agent.call`	read	Dispatch a task to a local agent on the same daemon (fresh session by default — replaces ad-hoc
`agentx`	read	AI-powered agentic code generation agent. Generates components, pages, APIs, documents, skills, and more for any tech stack.
`agentx_agents`	read	List all agents registered on the daemon with their status (active tasks, total tasks, errors, tier).
`agentx_approval`	read	Ask the operator for a yes/no decision by raising a decision card in their Approvals inbox, instead of asking in chat.
`agentx_attach_answer`	read	Explicitly answer the message currently claimed by this session (attach mode). Usually unnecessary — replying normally after agentx_attach_next is captured automatically. Use this only when your reply to the channel should differ from what you told the user in the terminal.
`agentx_call_owner`	read	Ring the owner for a live voice call on their desktop widget, when you need them directly and a message will not do.
`agentx_camera_ask`	read	Ask the owner to show you their phone camera, when you need to see something (a rack, a cable, a document).
`agentx_camera_look`	read	The newest picture from the owner
`agentx_channel_label`	destructive	Add and/or remove labels on a GitLab issue or merge_request through the canonical adapter (per-agent token, ledger). Use this instead of curl when changing labels on the entity that triggered the current task.
`agentx_crons`	read	List cron jobs and their health. Shows healthy, failing, disabled counts and per-job status with consecutive error counts.
`agentx_debug`	write	Toggle debug mode on the daemon. Enable verbose logging for specific categories (webhook, agent, channel, cron, mesh, context, memory, all) or disable it.
`agentx_detect_output_type`	read	Auto-detect the best output type for a given task description based on keyword analysis.
`agentx_events`	read	Read what happened on this node that matches your event subscriptions (agents.<id>.subscriptions in agentx.json): tasks other agents finished or failed, workflow runs, mesh hand-offs.
`agentx_generate`	read	Generate code, components, pages, APIs, documents, tests, workflows, schemas, emails, diagrams, and more using AI. Understands the project
`agentx_graph_review`	read	Triage pending intent-graph classifications via the configured review agent. The review agent sees each pending classification, may call
`agentx_health`	read	Get the daemon health status including node info, agents, crons, mesh peers, and uptime.
`agentx_inspect`	read	Analyze a project and return its tech stack, frameworks, databases, schemas, installed skills, and dependencies. Use this to understand a project before generating code.
`agentx_recent`	read	Read the most recent messages from a chat across ALL agents that have sessions for it. Returns inbound + each agent
`agentx_request`	read	Say what you are doing with a request the owner gave you, so it stays on their list of open requests until it is closed.
`agentx_schedule`	write	Manage recurring routines (schedules) in plain English. Actions: list, create, pause, resume, delete.
`agentx_send`	write	Low-level outbound send when you ALREADY have a channel-native chatId (Telegram numeric chat id, GitLab
`agentx_signal`	write	Stop a running agent task cleanly, or resume one that was stopped (#857).
`agentx_skill_match`	read	Find installed skills that are relevant to a given task description. Returns matched skills with relevance scores.
`agentx_task`	write	Send a task to a specific agent on the daemon. The agent processes it and returns a response. Use to delegate work to specialized agents.
`agentx_voice_queue`	read	See what this host is saying out loud and what waits to be said: every agent
`agentx_wiki_interview`	write	Run a scripted wiki interview — synthesize ONE typed article from a list of Q&A answers, and save it. Non-interactive: caller supplies the answers upfront. Use for capturing tacit knowledge (person/project/decision/pattern/...) that never hit a channel. For editing an existing article, use
`agentx_wiki_patch`	write	Edit a single wiki article via an LLM-applied instruction and write the result. Use this for targeted fixes, additions, or clarifications to an existing article when you already know what
`agentx_workflow`	write	Run a request that has several steps as a workflow, so AgentX follows it to the end instead of you chasing each step.
`api-service`	read	REST API service with auth, validation, error handling, and tests
`ask_user`	read	Ask the user a clarifying question when you need more information to proceed. Use this when the request is ambiguous or you need to confirm important decisions.
`billing-cutoff`	read	Invoices close on the 25th
`boom`	read	Always errors
`branching`	read	Classify, then route on the result via a branch node.
`channel.assign`	write	Set assignees on a GitLab issue or MR (replaces the assignee list). Pass empty array to unassign. Use during Triage→To Do flip to assign the coding agent, or to designate reviewer/merger on an MR.
`channel.create_issue`	write	Create a GitLab issue from an agent (PM test cases, CX client requests). Honors per-agent identity; resolves assignee usernames to ids; optional /relate to a parent issue.
`channel.label`	destructive	Add and/or remove labels on a GitLab issue/MR through the adapter (per-agent token, ledger, no raw curl).
`channel.reply`	write	Post a reply to a channel (gitlab/github comment, telegram/whatsapp message, ...) through the canonical adapter — auto-marker, auto-dedupe within 60s, identity resolved from agentMappings.
`chrome-extension`	read	Browser extension with popup, content script, background worker, and options
`cli-tool`	read	CLI application with commands, flags, interactive prompts, and config
`code`	read	Software work
`component-library`	read	UI component library with Storybook, tests, and documentation
`create_files`	write	Create one or more files as output. Use this when you need to generate code, documents, configs, or any file-based output.
`create_task`	write	Queue a new task for the user. Use when they ask you to build, fix, or work on something. CONFIRM the summary verbally first.
`data-pipeline`	read	Data processing pipeline with ingestion, transformation, and output stages
`deploy.staging`	write	Deploy
`drift.test`	read	test
`echo`	read	Echo back input
`edit_file`	write	Apply search-and-replace edits to an existing file. Use this for targeted modifications to existing code rather than rewriting entire files.
`extract`	read	Pull typed fields from free-form text via extract.structured.
`extract.structured`	read	Single-call typed extraction: prompt + JSON-Schema → structured data via forced tool_use
`f1`	read	d
`failure-0123456789`	read	beta: Bash keeps failing
`file.read_lines`	read	Read a text file as an array of lines (utf-8, 8MB cap, scoped to allowed roots)
`file.write_jsonl`	write	Append JSON records to a file as newline-delimited JSON (utf-8, scoped to allowed roots)
`fix.bug`	read	Fix a defect
`foo.bar_baz`	read	x
`fullstack-app`	read	Full-stack application with database, API, UI, and deployment config
`gadget-key`	read	Gadget admin key
`gadget-owner`	write	Gadget service is run by the platform team
`gadget-port`	write	Gadget service listens on port 7420 since the move
`get_credit_balance`	read	Get the user
`get_project`	read	Get a single project with its tasks and recent conversations. Use after list_projects when the user picks one.
`http.fetch`	read	GET a URL, return body + status + headers (1MB cap, http/https only)
`http.post`	write	POST a JSON body to a URL, return body + status + headers (1MB cap, http/https only)
`infra-x`	write	deploy and ssh notes
`initech-broken`	read	broken
`initech-cx-email`	read	email skill
`initech-cx-email-v2`	read	email skill
`initech-stale`	read	infra
`initech-v2-deploy`	write	Deploy Initech V2
`linear`	write	Trigger → Agent → End. Smallest workflow that runs.
`list_attachments`	read	List files the user has uploaded to the workspace. Filter by project_id and/or task_id.
`list_directory`	read	List files and directories at a given path. Use this to explore project structure.
`list_projects`	read	List the example.com user
`mesh.delegate`	write	Send a task to a remote mesh peer
`mixed-subjects`	read	12 turns over three unrelated subjects with genuine follow-ups inside each.
`mobile-app`	read	Mobile application with navigation, screens, and native features
`no-mock-db`	read	Use a real database in tests
`ops`	read	Infrastructure
`p1`	read	d
`p2`	read	d2
`parking`	read	Parking
`pick`	read	d
`ping`	read	HTTP echo
`ports`	read	d
`pref`	read	terse
`r1`	read	d
`rag.lexical`	read	BM25-style lexical search over an agent
`read_file`	read	Read the contents of a file. Use this to inspect existing code, understand patterns, check implementations, or gather context before generating code.
`release-follow-up`	write	Deploy, check it is live, tell the client (with your approval). Followed to the end; one summary for you.
`report`	read	Report what is visible and how it bears on the question
`respond_with_structure`	read	Return a structured value matching the schema
`restart.service`	write	Restart
`retry`	read	Per-node retry policy with a branch fallback path.
`review.merge-request`	write	Review an MR
`role`	read	Alex is a solo founder — prefers terse tech replies.
`run_command`	write	Execute a shell command. Use this to run build tools, test commands, linters, or inspect the environment. Commands run in the project root directory.
`saas-starter`	read	Full-stack SaaS starter with auth, billing, dashboard, and landing page
`search_files`	read	Search for files matching a pattern and optionally search their content with a regex. Use this to find relevant code, understand project structure, or locate specific patterns.
`server_ports`	read	daemon ports
`signal.resume`	write	Resume a task a stop signal paused, from its resume plan, in the same chat
`single-thread`	read	8 turns on one continuous task. Every turn after the first depends on the last.
`slow.thing`	read	test
`spawn_agent`	read	Spawn a sub-agent to handle a task in the background. The sub-agent runs in its own session and returns results. Use this for parallel research, long-running tasks, or delegating work to a specialized agent.
`terse-switches`	read	10 short turns with unsignposted subject changes and one-word follow-ups.
`test.capture`	read	test
`test.echo`	read	test
`test.strict`	read	test
`tone`	read	Gadget replies stay short
`u1`	read	d
`unnamed`	read	No description
`update_task`	write	Update a task
`wacli`	read	d
`x`	read	d
`x.y`	read	v1
`y`	read	d

Trust audit

BLOCKgrade F · trust 32/100 Do not install this without reading the findings. The audit found something that could harm you or your machine.

Layer	What it checks	Result
L0	Provenance & inventory	FAIL
L1	Static analysis of the code	FAIL
L2	Instruction surface (what it tells the agent)	PASS
L3	Class-specific surface	WARN
L4	Behavioural (sandbox)	SKIPPED

What the source does

Filesystem: declared (9 observation(s))
Network: declared (7 observation(s))
Shell: declared (14 observation(s))
Dependencies: not all pinned
Secrets in source: found

Findings (25)

HIGHInventory / provenance · inv.suspicious_name · CWE-1104

src/crons/event-payload.ts

event-payload.ts

Why it matters. member named after an attack tool

Fix. remove or justify