Agentburn

Name: Agentburn
Author: Socialpranker

by Socialprankercommunityv0.11.0

agentburn is a local, read-only CLI and MCP server that analyzes AI assistant log/database data to explain where token and cost burn comes from, including source breakdowns, overnight usage, loops, retries, and subagent activity. It supports assistants like Hermes Agent, OpenClaw, and Claude Code and can also generate fix recommendations, drift comparisons, and shareable burn reports.

From the registry: Local profiler for AI agents: burn by source, overnight bill, behavioral forensics, config fixes

GitHub Repository Website

Install & configure

Prerequisites

optional
Install uv
Install the uv Python package manager/runner to launch agentburn via uvx, which is the simplest portable path shown in the README.
```
$ install uv
```
https://docs.astral.sh/uv/
optional
Or install agentburn
Alternatively install the agentburn CLI so the `agentburn` command is available directly.
```
$ pip install agentburn
```
https://pypi.org/project/agentburn/

Install in your client

Paste into the AI you're chatting with — it will figure out the rest

Please install the `agentburn` MCP server into my current AI client (that's you).

Optional prerequisites:
- Install uv — Install the uv Python package manager/runner to launch agentburn via uvx, which is the simplest portable path shown in the README. Run: `install uv` (https://docs.astral.sh/uv/)
- Or install agentburn — Alternatively install the agentburn CLI so the `agentburn` command is available directly. Run: `pip install agentburn` (https://pypi.org/project/agentburn/)

Canonical MCP server config (stdio transport):
- command: `uvx`
- args: ["agentburn","mcp"]

Note: README states the MCP server is a zero-dependency stdio server exposed via `agentburn mcp`. It also shows a direct CLI form using command `agentburn` with args `["mcp"]`; the more portable path selected here is `uvx agentburn mcp`. No environment variables are documented for MCP startup.

Add this MCP server to my current client's config in the correct format for you. If you need secrets or credentials I haven't provided, ASK me — do not invent values or leave raw placeholders. After adding it, tell me how to verify the server is connected.

Note: README states the MCP server is a zero-dependency stdio server exposed via `agentburn mcp`. It also shows a direct CLI form using command `agentburn` with args `["mcp"]`; the more portable path selected here is `uvx agentburn mcp`. No environment variables are documented for MCP startup.

Tools (7)

agentburn

Runs the main cost/burn report for agents on the machine, showing where spending goes by source and time period.

agentburn why

Performs behavioral forensics on logs, including loops, retry storms, failed runs, idle heartbeats, and per-source breakdowns.

agentburn fix

Produces recommended configuration changes with estimated monthly savings.

agentburn drift

Compares the user's model spending against public 4-week model usage trends and flags drift.

agentburn rank

Compares the user's efficiency metrics against anonymous community percentiles (Burn Index).

agentburn mcp

Installs or exposes the MCP/server interface so an assistant can answer questions about its own spending.

agentburn explain

Uses a local or optionally remote LLM to explain the report in plain language.

Packages

agentburn

pypiv0.11.0transport: stdio

Details

Language: Python
License: MIT
Released: Jun 10, 2026
Last commit: Jun 22, 2026
Registry name: io.github.Socialpranker/agentburn

Related Servers

Claude Flow

AI orchestration with hive-mind swarms, neural networks, and 87 MCP tools for enterprise dev.

66.0k

Codebase Memory Mcp

Persistent codebase knowledge graph. Survives session restarts and context compaction.

35.4k

Serena

A powerful toolkit for coding, providing semantic retrieval and editing capabilities.

26.9k