agent-stack

Six small, single-concern reliability libraries for production LLM agents — TypeScript, Python, and MCP servers. Zero runtime dependencies. Adopt one or all.

Landing page: https://mukundakatta.github.io/agent-stack/ Paper: Six Reliability Primitives for LLM Agents: An Artifact Pattern for Stackable, Single-Concern Libraries HuggingFace Hub: mukunda1729/agent-stack

The six primitives

Library	Concern
AgentFit	Token-aware context-window fitting
AgentGuard	Network egress allowlist for tool calls
AgentSnap	Snapshot tests for tool-call traces
AgentVet	Tool-arg validation with retry hints
AgentCast	Structured-output validate-and-retry for LLM JSON
AgentBudget	Per-run token + dollar caps

Each ships in three forms: TypeScript on npm, Python on PyPI, MCP-server variant.

Install

# TypeScript
npm i @mukundakatta/agentfit @mukundakatta/agentguard @mukundakatta/agentsnap \
       @mukundakatta/agentvet @mukundakatta/agentcast @mukundakatta/agentbudget

# Python
pip install agentfit agentguard agentsnap agentvet agentcast agentbudget

// MCP server (Claude Desktop config)
{
  "mcpServers": {
    "agentvet": { "command": "npx", "args": ["-y", "@mukundakatta/agentvet-mcp"] },
    "agentguard": { "command": "npx", "args": ["-y", "@mukundakatta/agentguard-mcp"] }
  }
}

Read more

• 📖 Medium
• ⚙️ dev.to
• ✍️ Hashnode
• 📝 Paper (Zenodo)

Find them all

GitHub topic search across the family: topic:agent-stack on user MukundaKatta — 24 repos pinned.

Citation

@misc{katta2026agentstack,
  author    = {Katta, Mukunda Rao},
  title     = {Six Reliability Primitives for LLM Agents:
               An Artifact Pattern for Stackable, Single-Concern Libraries},
  year      = {2026},
  publisher = {Zenodo},
  doi       = {10.5281/zenodo.20074702},
  url       = {https://doi.org/10.5281/zenodo.20074702}
}

Local dev

To preview the landing page locally:

python3 -m http.server 8000
# open http://localhost:8000

License

MIT.