Skip to content

MCP tools

The tools a company's MCP endpoint offers to an AI client, and the level each needs. Work with Tinfirm from your AI client explains how to connect one and choose its level.

Every tool works in the company of the endpoint, as you, with the same rules as the web app: what you cannot see there is "not found" here. A client sees every tool in its list, and a call beyond its level is refused.

Your work

Tool What it does Level
get_inbox What waits for you: your decisions, and your conversations that wait on an account, failed, or wait on a runner host or your budget Read
list_agents The agents you may use, with their handles and status Read
get_agent One agent, by id or @handle, and whether it is ready for you: who would pay, which account it would use on each system, and whether a runner host can run it Read
list_conversations Your conversations, newest activity first Read
get_conversation One of your conversations: its accounts, cost, latest result or failure, open decisions, outputs and newest messages Read
list_decisions The questions and approvals of your conversations, newest first Read
get_decision One decision: its question and options, or the call it approves, and its status Read
start_conversation Start a conversation with an agent, your message its first request Read and write
send_message Send your message in one of your conversations, optionally as the answer to its open questions Read and write
choose_account Pick the account a waiting conversation should use on a system Read and write
cancel_turn Cancel a conversation's running or queued turn and its open decisions; the conversation stays open Read and write
retry_turn Run a conversation's failed or cancelled last turn again Read and write
end_conversation End a conversation for good Read and write
decide Answer a question or decide an approval of yours. A team's approvals are decided in the web app only. Read, write and decide

Memory notes

Tool What it does Level
list_memory_notes The notes an agent keeps for you, and whether memory is on for you Read
delete_memory_note Delete one of your notes, by its number Read and write
list_trigger_memory_notes A trigger's notes and the notes its conversations proposed, for its author, its team, Owners and Admins. Proposals are accepted or rejected in the web app only. Read

How the company is set up

Tool What it does Level
list_triggers The triggers you see: their source, system, agent, status and last firing Read
get_trigger One trigger as its page shows it: its events and conditions, its agent and instruction, the accounts its runs use, who pays, its team and its status Read
list_trigger_firings A trigger's firings, newest first, and for each one whether it started a turn or why it did not Read
list_systems The systems you see, and your own MCP servers Read
get_system One system's settings, never a secret Read
list_system_tools A system's tools, whether each reads or writes, and whether it is enabled Read
list_webhook_deliveries A webhook system's recent deliveries and what each fired. Owners and Admins. Read
try_webhook_delivery Show how a webhook system would read a body, and which triggers it would fire, without storing or firing anything. Owners and Admins. Read
test_system Test whether Tinfirm reaches a system you manage, as its Test button does Read and write
get_spend The spend you may see, as the Spend page shows it Read

Limits

  • Requests: each token or connected client may send 120 requests a minute, unless the operator changed the limit. Beyond it, the endpoint answers that the client must wait.
  • New conversations: at most 30 an hour per person and company start from AI clients, the Outlook add-in included, unless the operator changed the limit.
  • Results: a result's text for the model is cut at 40,000 characters; the full result is in its structured content.
  • Changes: no tool changes agents, settings, systems, triggers, accounts, members, secrets or runner hosts, and no tool returns a secret or uploads a file. test_system only records the system's last test, as the Test button does.