Resources Reference
The three read-only JSON resources the MCP server publishes so an agent can ground itself before it calls a tool.
MCP resources are read-only JSON snapshots the server exposes over a URI scheme. Agents load them to ground themselves before running tool calls: reading the ability catalogue, for example, lets an agent propose the minimum-privilege token to mint for a given task, and reading the error catalogue lets it pick the right remediation before a retry.
URI scheme
Every resource URI starts with zusegenius://. Three resources are published:
| URI | Contents |
|---|---|
zusegenius://enums/ability | The full ability catalogue a personal access token can carry: the entity:action gates that mirror the team permissions, plus the extras that exist only for tokens and agents (account:read, account:write, mcp:use, ci:deploy, ci:test-results). |
zusegenius://enums/error-code | Every error code the REST API and the tools can emit, with HTTP status, default message, remediation hint and docs_url anchor. |
zusegenius://enums/webhook-event | The outbound webhook catalogue grouped by family, with the minimum ability a subscribing token must hold for each event and the keys its data object can carry. |
Why load them
An agent that knows the catalogues before making calls will:
- Know what the current token can and cannot do, and ask for the minimum-privilege ability list when a new token is minted.
- Pattern-match an error code to its remediation without a docs round trip: a baseline lock needs a change request, a refused transition names the statuses allowed, an exhausted AI budget waits for an administrator.
- Pre-filter webhook events when registering an endpoint with
create_webhook_endpoint, so the list lines up with what the source actions produce and the token may receive.
Example
Reference a resource in a prompt:
Load
zusegenius://enums/webhook-eventand tell me which project events a token withproject:viewmay subscribe to.
Claude fetches the resource, parses the JSON, and answers without calling any tools.
Payload shape
There is no shared envelope. Each resource returns JSON keyed by its own collection name, with whatever counts are meaningful for it, and none of them carry a version field.
| Resource | Top-level keys |
|---|---|
zusegenius://enums/ability | abilities[], core_count, extras_count, total |
zusegenius://enums/error-code | error_codes[] |
zusegenius://enums/webhook-event | webhook_events[], families[] |
{
"abilities": [
{
"value": "project:view-any",
"group": "project",
"label": "View Any Project"
}
],
"core_count": 60,
"extras_count": 7,
"total": 67
}Each webhook event row carries name, family, required_ability and payload_keys; each error code row carries code, http_status, message, remediation and docs_url.
The JSON manifests the API host serves at /abilities.json,
/error-codes.json and /webhook-events.json are separate documents with
their own envelope. If you are switching an agent between the two sources,
check the shape rather than assuming it.
Prompts
The MCP server does not publish any prompts today; the full surface is tools plus the three resources above. The agent playbooks (implement a story from its context pack, write its tests, triage a defect) arrive as MCP prompts with the planning and test modules.
How is this guide?
Last updated on