PlugWiki
Agentistics logo

Agentistics MCP Server: Usage Stats and Task Board for Coding Agents

OfficialLocal
Getnpm@agentistics/mcpagentistics/agentistics

Lets Claude Code and other assistants read your AI coding usage and cost, and run a shared task board and session groups.

This is the MCP side of agentistics, an app that reads the session files your coding assistants write on your computer. It does not work alone. The tools forward requests to a running agentistics server, which is on port 47291 by default. The agentistics installer sets that server up and adds this MCP server to Claude Code for you.

Beyond usage numbers, it can create and edit tasks, send messages between sessions, and move a session to another account. Those tools are meant for agents that manage other agents, so read the safety section before you give an assistant all 48.

Good for

  • People who use several coding assistants and want to ask which one costs most
  • Developers running a fleet of agent sessions that need a shared task board
  • Teams that already run the agentistics dashboard and want it inside the assistant

Works in

12 should work

Tap an app to see how to set it up and any known problems.

Install Agentistics MCP

Claude Code

  1. 1

    Copy this prompt.

    Add the Agentistics MCP server to Claude Code. Run this command:
    
    curl -fsSL https://agentop.openvibes.tech/cli | sh
    
    Then check that it loaded and tell me if I need to sign in or restart.
  2. 2

    Paste it into Claude Code. It shows the change and asks before running it.

Requires Linux, macOS or WSL (README).

Run the agentistics installer first. The agentop server it starts adds this server to Claude Code for you at user scope, so the first method needs no extra step. The npx methods only work while that server is running, on port 47291 unless you set AGENTISTICS_API.

From the project README. Source.

Tools

  • Shows your total AI coding usage and cost

    Returns all-time totals across the coding assistants agentistics tracks: tokens, estimated cost, sessions, streak, most used model and top project. You can narrow it to one assistant.

    Try: “How many tokens and dollars have I spent on coding assistants overall?”

    Read · agentistics_summary

  • Compares your coding assistants side by side

    Lists each tracked assistant (Claude Code, Codex, Gemini, Copilot and others) with its sessions, messages, tokens, estimated cost and last active date.

    Try: “Which of my coding assistants costs me the most?”

    Read · agentistics_harnesses

  • Breaks cost down by model

    Gives token counts (input, output, cache read and write) and an estimated dollar cost for each model, for all assistants or one.

    Try: “Show cost per model for Claude Code only”

    Read · agentistics_costs

  • Lists your projects with usage and cost

    Returns each project with its session count, messages, tokens, estimated cost and last active date. You can scope it to one assistant.

    Try: “Which project used the most tokens this month?”

    Read · agentistics_projects

  • Lists your git repos with usage and cost

    Groups usage by normalised git remote, so the same repo on different machines or paths is counted once. Sessions with no linked repo go under "unlinked".

    Try: “Which repositories cost me the most?”

    Read · agentistics_repos

  • Shows your most recent sessions

    Returns the latest sessions with assistant, project path, duration, message count, tokens and model. First prompts in the result are the user's own text, so treat them as untrusted input.

    Try: “List my last 10 sessions and their cost”

    Read · agentistics_sessions

  • Checks your subscription limits before starting work

    Returns subscription budgets, pacing and a routing recommendation for which assistant and model to use next. Its description tells the agent to read it before opening sessions and not to start new workers on a plan that is on hold.

    Try: “Which assistant has room on its plan right now?”

    Read · agentistics_plan_limits

  • Lists your saved tags with totals

    Lists tags (saved filters over repos, projects, machines, teams or accounts) with sessions, cost and tokens added up for each.

    Try: “What tags do I have and what did each cost?”

    Read · agentistics_tags

  • Opens one tag in detail

    Returns one tag by name or id with totals, breakdowns by project, model, harness, repo and member, a daily activity series and the date window.

    Try: “Break down the "client-work" tag by model”

    Read · agentistics_tag_detail

  • Lists the widgets you can add to a dashboard layout

    Returns a fixed list of dashboard components with ids, categories, descriptions and default sizes. It calls nothing else. Use it before building or editing a layout.

    Try: “What chart widgets can I put on my custom dashboard?”

    Read · agentistics_component_catalog

  • Shows your saved dashboard layouts

    Returns each custom layout with its components, which one is active, and the instance id of each component so one can be removed.

    Try: “Show my dashboard layouts”

    Read · agentistics_get_layouts

  • Makes a new empty dashboard layout

    Adds a named, empty layout to the custom dashboard page.

    Try: “Create a layout called "Weekly review"”

    Write · agentistics_create_layout

Read from the source code

Permissions

high risk · 5 things to watch

High risk

The assistant can delete tasks, write into your other agent sessions and change dashboard settings in your agentistics server. Use it with a trusted local server and decline delete and session-message calls you did not ask for.

What it can do

  • Read your data
  • Create and change things
  • Delete things
  • Read-only mode

Watch out for

  • It can delete tasks and their history

    agentistics_task_delete removes a task with its comments, subtasks, files and links. Sessions filed under it are kept. Decline that call unless you meant it. unpkg.com

  • Agents can type into your other sessions

    agentistics_session_message writes into another session's conversation, and agentistics_session_switch_account copies a conversation to another account. Review those calls before you allow them. unpkg.com

  • Away mode can pre-approve releases and merges

    agentistics_session_away records what the owner pre-approved while away, such as releases, workers and merges, and tells new sessions about it. The server accepts writes only from a verified leader session. unpkg.com

  • Every tool call is logged to a file

    The server appends a record of each call (tool name, risk class, targets, a hash of the arguments, client and outcome) to agent-audit.jsonl in ~/.agentistics, or in AGENTISTICS_DIR if set. The file is created with owner-only permissions. unpkg.com

  • Task text and prompts are untrusted input

    Tool descriptions say first prompts and task comments may be untrusted text. Another session or person can write instructions there that your assistant then reads. unpkg.com

Security advisories

No published security advisories as of Nov 1, 2026.

Technical details
Turning on read-onlyNo read-only switch is documented or in the package. 21 tools only read, but every tool is advertised.
Talks toOnly the agentistics server you point it at, http://localhost:47291 by default
Sign-in and keysNo key or sign-in. The tools trust whatever agentistics server answers at the address in AGENTISTICS_API.

Questions

Is the Agentistics MCP server official?

Yes. It is published from the agentistics project's own repository and is the MCP side of the agentistics app.

Does it work without the agentistics server?

No. It forwards every request to a running agentistics server, on port 47291 by default. Start the server first, or set AGENTISTICS_API to its address.

Which assistants does it work with?

Claude Code is the main target, and the agentistics server registers it there for you. The package README also gives a Claude Desktop config. The agentistics docs say the server adds the MCP to Codex, Gemini and Copilot CLIs too.

Does it need an API key?

No. It sends no credentials. It only needs the address of your agentistics server.

Can I run it read-only?

No. No read-only setting exists in the package. The tools that change things are the task, session group, layout and session tools.

Does it send my data anywhere?

The MCP server talks only to the agentistics server address you give it. The agentistics README says that on a single machine the app opens no outbound connection, and in Team Mode a machine sends computed metrics only.

Why does the package README say 13 tools?

It is out of date. The published package registers 48 tools; two of them (the flow tools) stay hidden unless the server reports flows as on.

What is the license?

FSL-1.1-ALv2 according to the package metadata, which is source-available and not open source. The npm package does not ship a license file.

Information

updated Nov 2, 2026

Publisher
Official
Package
@agentistics/mcp
Registry
npm
Transport
stdio
Auth
None
Version
2.173.1
License
FSL-1.1-ALv2
Last commit
Oct 31, 2026
Open issues
65

WebsiteGitHubnpmChangelogIssues

Sources

Docs, issue trackers and discussions this page is built from. Issues, quotes and findings above link to their own source.

Show all 4
  1. @agentistics/mcp README · github.com
  2. agentistics README · github.com
  3. agentistics MCP docs · github.com
  4. @agentistics/mcp 2.190.0 dist · unpkg.com

Facts from the official MCP registry, GitHub, npm, PyPI and the Docker catalog. Updated Nov 2, 2026.