AGENTS.md

Guidance for coding agents working on LMSSEP, the modular MCP server suite.

Project Overview

LMSSEP contains independent MCP servers for platform access, memory, planning, reasoning, reporting, monitoring, prediction, security, persona response, and prompt refinement. Each server is designed to be loaded only when needed to control token and tool surface costs.

Important paths:

README.md: suite overview and server list.
requirements.txt: shared Python dependencies.
/README.md: server-specific usage.
/main.py: MCP server entrypoint for most modules.
/requirements.txt: server-specific dependencies.
mcp_server/server.py: DDD chat backend entrypoint.
cli_proxy/: Copilot CLI subprocess wrapper.
predicting/: super-forecasting workflow.
reasoning/ and reasoning subfolders: structured reasoning servers.

Setup Commands

bashcd <project-root>
python3 -m venv .venv
source .venv/bin/activate
pip install -r requirements.txt
cd <server>
pip install -r requirements.txt
python3 main.py

Use <server>, <MCP_CONFIG>, and <API_TOKEN> placeholders in docs. Do not publish local machine paths or real tokens.

Server Boundary Rules

Keep each MCP server independently runnable.
Do not make one server import runtime secrets or config from another server.
Use configuration templates for external services.
Prefer loading only the required server for a task.
Put output under each server's documented output directory.

Configuration

Copy template config files before editing local credentials.
Keep Slack, Jira, Confluence, API keys, and LLM provider keys out of source control.
For examples, show placeholder config keys rather than real values.

Testing

Smoke test the specific server whose tool definitions changed.
Exercise tool schema changes with small local calls before connecting to an agent runtime.
Check that startup failure messages name missing config without printing secret values.

Security

Never commit configured credential files, output containing private messages, or ChromaDB memory dumps.
Treat platform and chat exports as sensitive.
Do not run security or monitoring tools against systems without explicit authorization.

LMSSEP MCP Suite Agent Guide