ClaudeMap

LLM Functions

Bash / JS / Python functions exposed as LLM tools, with two-way MCP support — a runtime layer for LLM agents.

Skills & Commandsskillstoolsbashmcp
Visit projectsigoden/llm-functions

sigoden's llm-functions turns Bash, JavaScript, and Python scripts into a uniform set of LLM-callable tools, with function-calling declarations autogenerated from comments. A bundled MCP server exposes your tools over the Model Context Protocol, and an MCP bridge lets external MCP tools be used inside llm-functions. Useful as a glue layer when you want an agent to reach local utilities without writing a custom MCP server for each one.

What it does

  • Write tools in Bash, JavaScript, or Python — function-calling JSON declarations are autogenerated from the comments you already wrote (argc-style annotations, JSDoc, or Python docstrings)
  • Agents follow the formula prompt + tools + documents (RAG), defined by a folder with an index.yaml — the README likens it to OpenAI's GPTs
  • Two-way MCP: mcp/server exposes llm-functions tools and agents over MCP; mcp/bridge pulls external MCP tools in
  • Prerequisites are just argc and jq; a web_search slot is filled by linking one provider-specific tool (Cohere, Perplexity, Tavily, VertexAI)

When to use it

Give a CLI agent (AIChat) a working set of local utilities — command runners, file processors, API calls — without hand-writing JSON schemas per tool; or make your existing scripts reachable from any MCP client by running the bundled server.

Getting started

Prerequisites: argc and jq. Then:

git clone https://github.com/sigoden/llm-functions
cd llm-functions

# list tool filenames in tools.txt, agent names in agents.txt, then:
argc build
argc check

# expose to AIChat (the only supported host):
argc link-to-aichat
# or point the AICHAT_FUNCTIONS_DIR env var at the repo

Things to know

AIChat is currently the only CLI that hosts llm-functions — the README says so explicitly, so plan around that dependency. Bash tools run with shell privileges (the demo execute_command.sh is an eval wrapper), so curate the toolset like you would for any shell-exposing agent. argc check exists precisely to verify env vars and Node/Python dependencies before first use.

Repository

sigoden/llm-functions

Charted

Related on the map

  • SureForge

    Skill

    An instruction-only skill that makes agents research before they ask, plan before they build, and verify before they call anything done.

    • workflow
    • verification
    • quality
    • skill
    View

  • guizang-product-video-skill

    Skill

    Guizang's agent skill turns your repo's recent updates into a finished promo video — real components, code-rendered animation, an original code-written soundtrack — no video editor involved.

    • agent-skill
    • video
    • promo
    • marketing
    View

  • geo-sleuth

    Skill

    An agent skill that geolocates a photo with nothing readable in it — OSM geometry, elevation skylines and satellite tiles turn “where was this taken?” into a measured answer with an error radius.

    • agent-skill
    • geolocation
    • osint
    • openstreetmap
    View