Skip to main content

Adapters

Each adapter owns its provider's behaviour, native contracts, specification and fixtures (Adapters and ownership). A native contract or a typed model is not an installed capability: the runtime column says what runs.

Runtime available​

AdapterRuns asOperations
GitLabcatalog provider18 reads and 4 approved writes from the pinned OpenAPI document, and the two merge-request feed reads
Catalog providerconnectors-catalog-providerJira, Confluence, HubSpot, Zendesk, Google, Runpod and Slack, from pinned API documents
Kubernetesconnectors-kubernetesresources.list, endpoints.discover, optionally hosts.discover and the Helm release reads
SQL (PostgreSQL and MySQL)connectors-sqlquery.read, schema.list, database.list, table.list, table.describe, index.list
Tavilyconnectors-tavilywebsearch.search, websearch.fetch, websearch.crawl
Lokiconnectors-lokilogs.query_range, logs.query_metric, logs.labels, directly or through Grafana's data-source proxy
Prometheusconnectors-prometheusseries.query, series.query_range, rules.list, directly or through Grafana's data-source proxy
Grafanaconnectors-grafanadatasources.list; discovery and the mediated route are specified only

Specification or design only​

AdapterWhat exists
Atlassianan ESS model of the native document vocabulary for Jira and Confluence; Jira and Confluence reads run through the catalog provider
Alertmanagera design

MCP has specified invocation, projection, auth lifecycle, mutation replay and composition contracts in the repository's adapters/mcp/ directory and no runtime. Slack and WebRTC have no native adapter documentation; Slack conversations are read, and messages posted, through the catalog provider.