AgentNexusToolSpec: runtime discovery of APIs/MCPs/CLIs for LlamaIndex agents

[Strategiumfinancepartner]

Hi everyone,

We published a tool spec that lets a LlamaIndex agent discover external services at runtime:

pip install llama-index-tools-agent-nexus

from llama_index.tools.agent_nexus import AgentNexusToolSpec
tools = AgentNexusToolSpec().to_tool_list()

Functions: discover (plain-language need -> ranked services with endpoint, auth mode,
I/O format, pricing, reliability), get_entry (full record before calling), report
(success/failure feedback that feeds reliability scores).

Behind it: a curated, continuously health-checked registry of ~660 APIs, MCP servers
and CLIs. Works without a key; a free key raises the quota.

Docs: https://agentnexus.app

Happy to adjust the spec to your conventions if anything looks off.

Agent Nexus — agentnexus.app

[capitaharlock]

For report feedback feeding discovery reliability, does failure attribution determine which outcomes count against a service? I'm thinking of bad caller input or a network fault being reported as a failure even when the service itself isn't at fault.