Three rules that decide everything
An agent is a person's hand, not a fourth party.
It connects as one existing account, with a token that account issued, and has that account's permissions minus the signatures below. Never more. No agent-only identities, no service accounts.
Agents propose; people sign.
Approving another person's hours, confirming a contract, sending an invitation, sending an invoice: these are human signatures and are not available to agents. Submitting your own hours is the one signature an agent may trigger, because the record is yours and your token authorised it.
The record says who acted.
Every write made through an agent is stamped as made via that agent, on the security trail and inside the signed audit record. The trail can always say: the agent proposed, the person signed.
What an agent can do here
The routine work, in plain terms. Each action runs through the same rules a person meets on the same screen, so an agent placing hours is refused by the same weekly cap, in the same words, that would refuse you.
Working for a professional
- Check the week: declared hours, committed blocks, cap and what is left
- Place hours for accepted work, inside your capacity and caps
- Log the hours you worked and submit them in your name
- Read the work requests waiting for you, and accept or decline when you say so
- Draft the month's invoice from approved hours
- Read your standing and the terms of any contract you hold
Working for a business
- See the bench: who is available, and their standing
- Compose and send a work request under a contract, with the same budget thresholds and caps a person meets
- Track the workspace's requests by state
- Read the hours submitted and awaiting approval, with anything flagged
- Read active contracts, their terms and budgets, and this week's standing against them
What an agent can never do
These stay with a person, on purpose. They are the signatures a counterparty relies on, and a record is only worth checking if a human put their name to it.
- Approve or dispute another person's hours. Verified is a human countersignature; an agent can get a record to submitted, never past it.
- Draft, send, confirm or end a contract.
- Send invitations of any kind, or manage workspace members.
- Send an invoice to a client, or change its status.
- Export the signed records. Those are for people and auditors; an agent can read the summary.
Why it is safe to say yes
- Scoped tokens. You choose what each agent may read and write, area by area: schedule, time, requests, billing, standing. A missing scope returns a clear refusal, not a workaround.
- Revocable in one click. The Integrations page shows every token, its calls and its last use. Revoke it and the next call fails.
- Two-factor to issue. A token can only be created by an account with confirmed two-factor authentication, from a session that just passed it. A staff reset of your two-factor revokes every token on the account, because the lost device may be where the agent runs.
- Expiry by default. Tokens expire in 90 days by default; a replacement is issued from the same page.
- Visible on the trail. Every write leaves a row: which tool, whether it was done or refused, which record it touched. Never the arguments. Your workspace's audit trail shows "via agent" next to the human it acted for.
Illustrative trail. The agent placed and drafted; the person submitted; the client verified. Nothing an agent did counts as verification.
How it connects
- Turn on two-factor authentication on your Tesserow account, if you have not already.
- Issue a token from Settings, Integrations: name the agent, tick the scopes it needs, copy the token once.
- Point your agent at the endpoint. Tesserow speaks MCP (Model Context Protocol) over HTTP at a single endpoint. Paste the endpoint and the bearer token into any MCP-capable client. No SDK, nothing to install on Tesserow's side. Building a client or reviewing the connector? The MCP reference lists the endpoint, scopes, every tool, and the limits.
- Start from a prompt. Two starter prompts ship with the server: Plan my week reads your planner and inbox and proposes blocks within your caps, reporting what was refused and why; Log yesterday turns a line of notes into draft records and submits them after you confirm.
The server's own instructions to every agent are the same as ours to you: propose before you place, draft before you submit, and confirm with the person before any write they have not seen.
Nothing about what you rely on. The hours you approve were still submitted under a named person's signature. What you gain is a trail that shows when an agent did the drafting, so "who actually did this" has an answer instead of a shrug.
Frequently asked questions
Can an AI agent approve hours or confirm a contract on Tesserow?
No. Approving another person's hours, confirming a contract, sending invitations and sending invoices are human signatures and are not available to agents. An agent can get a record as far as submitted; a named person takes it from there.
Which agents work with Tesserow?
Any client that speaks MCP and can hold a bearer token. Tesserow does not favour or name particular agent products; the endpoint is the same for all of them.
Does using an agent weaken the verified record?
No. The agent acts as your hand with your permissions, minus the human signatures. Every write is stamped via that agent, on the security trail and inside the signed audit record, so the record can always say the agent proposed and you signed.
Can I see what my agent did and stop it?
Yes. Integrations shows each token's calls and last use, and revokes it in one click. Revoked tokens fail on their next call.
Is there a cost?
Agent access is part of the account, not a separate product. Core tools for professionals are free; businesses pay for the coverage view and the audit trail, and Tesserow takes no cut of anyone's rates.
Bring your own agent. Keep your own signature.
Tesserow for agents is available to early-access accounts. Tell us which client you use and what you want it to handle.
Request early access