Skip to content

MCP

The tools cats mcp registers. Each probing tool is the matching cats run subcommand over a different transport, and returns the same JSON. The server's own instructions, sent to the agent on connect, describe the workflow; cats prompts agentic is the longer reference for an agent, and Agentic is the guide for the human wiring it up.

Tools

Tool cats run Returns Hints
cats_list_accounts every account: name, account_id, tags, declared targets reads config, no AWS call
cats_account_info account-info name, account_id, tags, deployed_version, targets read-only
cats_http_call http-call result, reason, http_status read-only; sends traffic to the host you name
cats_tcp_call tcp-call result, reason, http_status (always null) read-only; sends traffic to the host you name
cats_udp_call udp-call result, reason, correlation_id writes a receipt to S3 when delivered
cats_receive_s3 receive-s3 received read-only
cats_shell shell stdout, stderr, exit_code, duration_ms, stdout_truncated, stderr_truncated destructive: arbitrary commands on a listener; needs allow_shell: true at deploy
cats_write_profile write-profile profile_name, profile_file, role_arn, parent destructive: rewrites the user's ~/.aws/config (the only file it writes; the CLI's --profile-file has no MCP equivalent)
cats_validate cats validate valid, message, config_path reads config, no AWS call
cats_infra_status cats infra status installed_version, accounts read-only

The Hints column is each tool's MCP ToolAnnotations in words, so a host can apply its approval policy. "Sends traffic to the host you name" is the spec's openWorldHint: the tool reaches outside CATS even though it changes no CATS state. cats_shell and cats_write_profile are annotated so a host asks before each call.

result values for the probes are defined in Network Testing.

Tool reference

list_accounts(config_path=None)

account_info(account, config_path=None)

http_call(account, target, host, port, config_path=None)

tcp_call(account, target, host, port, config_path=None)

udp_call(account, target, host, port, receiving_account=None, config_path=None)

receive_s3(account, target, correlation_id, config_path=None)

shell(account, target, command, config_path=None)

Examples:

cats_shell(account="prod", target="priv", command=["dig +short example.com"]) cats_shell(account="prod", target="priv", command=["ss -tlnp | grep 8080"]) cats_shell(account="prod", target="priv", command=["cd /tmp", "echo hi > f.txt", "cat f.txt"])

write_profile(account, role='default', profile_name=None, config_path=None)

validate_tool(config_path=None)

infra_status_tool(account=None, config_path=None)