Agents and integration
Point an MCP client or a REST script at one endpoint. It authenticates with a key you created and scoped, never with a Meta token.
Authentication
Agent credentials look like tag_live_<prefix>_<secret>. The prefix locates the credential record; only an Argon2id hash of the secret is stored. Authenticating produces a principal containing the agent and workspace identifiers — not permissions asserted by the caller. For MCP clients that support the current authorization profile, Readyweek acts as an OAuth protected resource and issues its own audience-bound access tokens. Those are our tokens. A Meta token is never accepted as an agent credential.
Read tools
Nine read tools draw on the four read capabilities. Each requires the matching capability on the connection it names.
- threads_get_profile
- threads_get_recent_posts
- threads_get_post
- threads_get_post_insights
- threads_get_account_strategy_performance
- threads_get_replies
- threads_get_conversation
- threads_get_quota
- threads_get_permissions
threads_get_account_strategy_performance returns medians and sample sizes for a labelled group against the same account’s own baseline. It returns no verdict and no cause, and it carries the observational notice in its output.
Write tools
There are two: threads_create_post and threads_reply. A deployment can run in a read-only configuration, and when it does these two are not advertised at all rather than being advertised and then refused. Both take an idempotency key and a reason. Both return a disposition — executed, approval required, or denied with reason codes. Quote, repost and delete are not implemented and are not advertised.
What the agent cannot do
- It cannot choose which workspace or account it acts on. That comes from its credential.
- It cannot ask Readyweek to call an arbitrary URL. The provider host and paths are fixed.
- It cannot raise its own permissions. Text it read from Threads is treated as data here. A post that instructs your agent to ignore your limits can change what your agent drafts — that part is between you and your model — but it cannot change what our server allows, because the limits are evaluated here and not by the model.
- It cannot learn your rule values from a denial. Denials return stable reason codes, not the thresholds that produced them.