GLOSSARY, read 10 Nov 2026
36 terms, each with an example from a real server
- column-level lineageLineage that follows individual columns, showing which source columns feed which output column.
- context windowThe limit on how much text a model can read at once, including the conversation, tool definitions and every tool result.
- Cortex AnalystSnowflake's natural-language analytics service.
- data observability incidentA recorded problem found by monitoring, such as late data, a failed test or a volume change.
- Discovery APIThe dbt platform API that returns metadata about a project's models, sources, exposures, tests and runs.
- dynamic client registrationA step where an MCP client registers itself with an authorization server automatically, so nobody has to create a client ID by hand.
- Genie spaceA Databricks workspace for asking questions of chosen tables in natural language, set up with curated instructions.
- human confirmationA step where a person approves a tool call before it runs.
- impact analysisThe question of what breaks downstream if an asset changes or fails.
- input schemaA JSON Schema that states which arguments a tool accepts, their types and which are required or limited to set values.
- local MCP serverAn MCP server that runs on the engineer's own machine or infrastructure, usually started by the client over stdio.
- MCP clientThe part of an AI application that connects to MCP servers, lists their tools and sends the calls the model decides to make.
- MCP serverA program that exposes tools, resources or prompts to an AI application over the Model Context Protocol.
- mcp-remote bridgeAn open-source Node.js proxy that lets a client which only speaks stdio reach a remote MCP server.
- metadata graphA connected store of data assets and their relationships: tables, columns, dashboards, owners and the links between them.
- monitor as codeDefining a data monitor as configuration kept in a repository instead of clicking it together in a UI.
- OAuth 2.1 authorization for MCPThe sign-in flow the MCP specification defines for HTTP servers.
- persona scopingLimiting what an agent can see or do by the persona of the signed-in person, such as admin, member or guest.
- personal access tokenA long-lived secret a person creates in a tool to let a script or agent act as them.
- read-only toolA tool that changes nothing in the system it talks to.
- remote MCP serverAn MCP server the vendor hosts and the client reaches by URL.
- resourceRead-only context an MCP server exposes by address, such as a file or a database schema.
- row-level securityA rule that filters which rows a person may read in a table or model, often by role or department.
- semantic layer queryA query written against defined metrics and dimensions instead of raw tables.
- service accountA non-human identity with its own role, used by unattended agents and jobs.
- stdio transportA way to run an MCP server as a local subprocess.
- Streamable HTTPThe MCP transport for servers reached over a network.
- text-to-SQLA tool that turns a plain-language question into a SQL statement, usually using what it knows about the schema or project.
- tokenThe unit in which models count text, roughly a word fragment.
- toolA named function an MCP server offers to the model.
- tool annotationsOptional hints a server attaches to a tool, such as readOnlyHint, destructiveHint, idempotentHint and openWorldHint.
- tool definitionThe description of one tool that a server sends to the client: its unique name, a title, a text description and an input schema.
- tool-definition sizeThe amount of text a server's tool definitions take: names, descriptions and input schemas together.
- tools/listThe protocol request a client sends to ask a server which tools it offers.
- Unity Catalog function as a toolA function registered in Databricks Unity Catalog and presented to an agent as an MCP tool.
- write toolA tool that creates, changes or deletes something: a table row, a monitor, a report, a tag.
Last reviewed 10 Nov 2026.