# MCP check: create and retrieve a card

Executed on November 1, 2026 by the Stellary team, from Codex, against the hosted MCP service. Fictional data; no private identifiers or secrets are reproduced here.

## Scope

An agent identity already authorized for the project. This check does not cover client installation, fresh OAuth authorization, a Desktop runner, rejected writes, or multi-agent delivery. No successful Claude Code result is claimed: the second-client check stopped on an expired model-provider session before retrieving the card.

## Steps actually executed

1. `stellary_init`: initialize the session and select the available identity.
2. `list_projects`: retrieve the accessible project. Reuse its returned identifier rather than guessing one.
3. `get_next_mission` with `mode: start`: open a mission with an explicit objective.
4. `create_card`: create a check card stating that it contains fictional data and triggers no paid mission.
5. `get_card_details`: retrieve that card using its identifier and the mission identifier.

Observed result: creation succeeded; the retrieved title and description matched the submitted fields. The card was open, with no assignment, comment, or checklist. This is a bounded functional check, not a speed measurement or model comparison.

## Repeat in your own project

Create a card with a non-sensitive witness sentence. Ask the client to find it, quote that sentence, and stop. To test a write, explicitly authorize creation of a new fictional card and retrieve it in the interface. Compare fields, project, identity, and permissions; a model response alone does not prove storage.

If the model provider requires sign-in, restore that session before testing. If Stellary returns 401, reconnect MCP access. For a 403 response, check project and tool permissions rather than expanding access by default. Finish the mission with an accurate outcome and preserve verification limits.