agent-breadcrumbs
Cross-platform agent logging and observability
claude mcp add --transport stdio ejcho623-agent-breadcrumbs npx -y agent-breadcrumbs
How to use
Agent Breadcrumbs is a lightweight MCP server that centralizes and standardizes logging across multiple agents and clients. It provides a single log_work tool that aggregates work performed by different agents for one or many users, with support for custom schemas or built-in profiles. You can configure various sinks to store or forward logs, including local JSONL files, webhooks, or Postgres databases, and you can view the results via a simple dashboard. The server enforces a consistent payload format via MCP schemas, making it easy for clients to emit correctly structured log records without repeatedly explaining payload details.
To use it, run the MCP server using the npx-based quick-start or install it locally from the repository. You can run with defaults or specify a server-config.json to tailor schema and sink settings. The server supports predefined profiles (agent insights, delivery tracking, audit trail, knowledge capture) and lets you define your own logging structures if you need custom data models. For cron-driven or scheduled workflows, you can instruct agents to call log_work at regular intervals and include contextual information in each log entry. The included dashboard app lets you view and analyze the logged work, and the CLI/config examples show how to wire the MCP server into client configurations.
How to install
Prerequisites:
- Node.js and npm (or use npx to run without a global install)
- Access to the MCP repository or npm registry where agent-breadcrumbs is published
Install and run (quick start):
-
Install and run with defaults using npx: npx -y agent-breadcrumbs
-
If you want to specify a config file, use: npx -y agent-breadcrumbs --config /absolute/path/to/server-config.json
Install from repository (advanced):
- Clone the repository or navigate to the packages/mcp directory in the repo.
- Install dependencies: npm install
- Build (if required by the repo setup): npm run build:mcp
- Run the MCP server with a sample config: node packages/mcp/dist/index.js --config packages/mcp/examples/server-config.agent-insights.sample.json
Run the dashboard locally (optional): npm run build:dashboard npm run dev:dashboard -- --config apps/dashboard/examples/dashboard-config.sample.json
Prerequisites recap:
- Node.js and npm installed
- Access to a config file or the ability to pass inline config through environmental setup
- Optional: a sink (jsonl, webhook, postgres) configured per your deployment needs
Additional notes
Tips and common considerations:
- The default sink is jsonl to ~/.agent-breadcrumbs/logs.jsonl unless overridden by the server config.
- When using npx, you’re running the latest published tool version; for reproducible environments, pin to a specific version if available or install locally.
- If you enable a non-file sink (webhook or Postgres), ensure network access and credentials are correctly configured in the server-config.json.
- The MCP Config Model supports either a custom schema or a built-in schema_profile; avoid setting both simultaneously.
- The dashboard app is separate from the MCP server package; ensure you have the corresponding config file to visualize the logged data.
- When integrating with agents like Codex, Claude Desktop, or OpenClaw, provide the correct config path to ensure log_work calls are emitted in the expected format.
Related MCP Servers
ironcurtain
A secure* runtime for autonomous AI agents. Policy from plain-English constitutions. (*https://ironcurtain.dev)
mindbridge
MindBridge is an AI orchestration MCP server that lets any app talk to any LLM — OpenAI, Anthropic, DeepSeek, Ollama, and more — through a single unified API. Route queries, compare models, get second opinions, and build smarter multi-LLM workflows.
codemesh
The Self-Improving MCP Server - Agents write code to orchestrate multiple MCP servers with intelligent TypeScript execution and auto-augmentation
mcp-agent
Lightweight, focused utilities to manage connections and execute MCP tools with minimal integration effort. Use it to directly call tools or build simple agents within your current architecture.
mcp-web-search-tool
A MCP server providing real-time web search capabilities to any AI model.
productboard
A **complete** Model Context Protocol (MCP) server that provides comprehensive integration with the Productboard API. More than 40+ tools.