LoadTable

LoadTable is early software. These pages describe the current development build; see Status for what is missing.

Setting up agents

Agents can use LoadTable two ways: through its MCP server, or by running the loadtable CLI from their shell. Most setups use both: MCP for submitting and waiting, the CLI for reading long logs.

MCP server

The server is built into the CLI: loadtable mcp, over stdio.

Claude Code

claude mcp add loadtable -- ~/.local/bin/loadtable mcp

Codex (~/.codex/config.toml)

[mcp_servers.loadtable]
command = "/home/you/.local/bin/loadtable"
args = ["mcp"]

Other MCP clients take the same command and argument.

Tools

Tool What it does
submit queue one command; returns the job id at once
submit_steps queue a plan of named, dependent steps
run_suite queue a suite from the repo’s run policy
wait wait until a job ends or a timeout passes; returns its state and log tail
logs read a job’s log
get one job in full
list jobs, filtered by owner or unfinished only
status the machine now: slots, jobs, pressure, peers
estimate how long a command usually takes
prioritize change a waiting job’s queue class
cancel cancel your own job
audit the audit log

Every call takes a who: each agent should pass its own name. Cancelling a wait never touches the job, so an agent can wait in short pieces and keep its tool calls under the client’s timeout.

AGENTS.md / CLAUDE.md

Paste this into the repo so every agent, of any vendor, knows the rules:

Every build and test goes through LoadTable:
  loadtable run --who YOUR_AGENT --anywhere -- <command>
--anywhere picks a free machine, --peer NAME picks one,
--pin keeps the job on this machine (GUIs, dev servers).
Sessions with MCP can use the loadtable tools instead.
Check `loadtable list --active` before starting. If a call
times out, query the printed job id; never resubmit.
Name test targets: cargo test -p X --lib FILTER.

Habits that help