Reference for AI agents (plain text)

atbridge.ai is built for AI agents. So its whole API is published as flat plain text — an agent (or a coding assistant, or a script) can slurp the entire command and tool surface in one fetch, with no HTML to parse.

The files

/llms-mcp.txt — every MCP tool the atbridge.ai server exposes: name, description, each parameter (type · required · description), and whether it is a write (writes require ATBRIDGE_ALLOW_WRITES=1). This is exactly what an MCP client sees from tools/list.
/llms-cli.txt — every CLI command and flag (the same operations as MCP). JSON in, JSON out.
/llms-full.txt — a task→result capability guide (what atbridge.ai does, framed as Situation → Task → Action → Result), for grounding an agent on when to reach for which operation.
/llms.txt — the LLM-readable site map (entity summary + the canonical links), in the llmstxt.org format.

Why plain text

Rendered documentation is built for humans — navigation, styling, scripts. An agent that has to parse HTML spends tokens on markup and can miss content. The .txt reference is the opposite: pure API, nothing to strip, cheap to ingest, and easy to diff. It doubles as the public API contract.

It never drifts

/llms-cli.txt and /llms-mcp.txt are generated from the live product — the CLI file from the real atbridge command tree, the MCP file from the server’s tools/list — by atbridge reference --kind cli|mcp, regenerated on every build. They cannot fall out of sync with what the binary actually ships.

Get set up first

To actually call these tools you still connect atbridge.ai to your agent once — see Connect Claude Desktop, Cursor, or any MCP/CLI client. Then hand your agent the files above.

Was this page helpful?