Skip to Content
Tracira

MCP server

Connect AI agents to Tracira over the Model Context Protocol. Agents report their runs and hold risky actions for a person to approve.

Tracira runs an MCP server for AI agents that work on their own, like Grok Bot routines. The agent reports what each run did and holds anything risky until a person approves it in Tracira.

Everything an agent sends goes through the same pipeline as POST /api/logs: your rules, the review queue and notifications all work the same way.

Connect

SettingValue
URLhttps://www.tracira.com/api/mcp
TransportStreamable HTTP
HeaderAuthorization: Bearer YOUR_TOKEN

Copy your token from the Integrations page in your workspace. Use the www address: the bare domain redirects, and some clients drop the request body on a redirect.

Tools

check_decisions

Call at the start of every run. Returns what the person decided on items the agent held before (do it, do an edited version, redo it, or don't), and lists items still waiting. Each decision is returned once.

FieldRequiredDescription
routineYesThe routine's name, the same on every run.
agent_nameNoThe agent's name. Defaults to Grok Bot.

report_run

Call once at the end of every run.

FieldRequiredDescription
routineYesThe routine's name, the same on every run.
summaryYesPlain-language summary of what the run did.
runs_everyNohour, day, weekday, week or month. Turns on missed-run alerts.
detailsNoWhat triggered the run or what it looked at.
waiting_for_approvalNoUp to 20 actions the agent did not do because a person must approve them.
agent_nameNoThe agent's name. Defaults to Grok Bot.

Each item in waiting_for_approval:

FieldRequiredDescription
whatYesOne line saying what would happen, e.g. Send email to anna@acme.com: Invoice reminder.
contentYesThe full draft: the exact message, or the exact details of the action.
toNoWho it goes to.
replaces_idNoWhen redoing an item after "changes requested", the id of the item it replaces.

How it shows up in Tracira

  • The agent becomes a project (its agent_name), and each routine a task.
  • Every run is one output. Each held item is its own output waiting under Needs you.
  • Items held in the same run share a conversation, so you see them together.

On this page