Tekelili is an MCP providing a simple interface for MCPs to:

• "Sign In" as an Agent/Character, giving them a base prompt
  • For serious agents, this can provide unique expertises or roles to each agent
  • For roleplay characters, their personality and behavior
• Start a Scenario Session involving between 1-6 agents/characters that supplements the agent(s) with additional instructions
  • For serious agents, this could be a pair/group programming session, a code review session, a security review, an incident response, etc
  • For roleplay / fun stuff: Entertaining scenarios for characters to work through as they help you solve your problem
• List, search, and describe available characters and scenarios and the content packs they belong to

Tekelili is a code-execution style MCP, based on topics dicussed by Anthropic and Cloudflare in late 2025. However, instead of TypeScript Containers running on either local docker or cloud infrastructure, Tekelili utilizises the Yith MCP SDK to provide a lightweight, in-process, and performant but still secure Lua Sandbox for LLM code to run in. Most of the Lua API calls that LLMs are able to use in tekelili provide both a lightweight textual response, and the ability to get structured results, and use the lua print() function to output exactly the results they need.

This implementation of tekelili is still a work in progress. Other features not yet implemented in this version but that were in prior prototypes of tekelili, which I want to implement here are:

• per-project & global configuration & storage backends (no config/storage exists now)
• Support for user-supplied content packs, rather than the handful of built-in ones that ship with tekelili
• Session tracking & Sign-In creating a Shift Log, which LLMs can update throughout their "shift" so that future agents can
• Agent Growth: Allowing an LLM agent/character to append to its base system prompt as it develops new skills, accomplishments, or personality quirks