The engines and agents a Context Engineering Studio repository serves: which deploy directly, which render a format for you to take away, and which read it over MCP.
Portability targets
One context repository serves several engines and agents. The Portability tab lists each target as its own card, and you build or deploy each one separately.
CES generates each engine-specific format when you ask for it, from the same canonical model. The repository itself stores nothing engine-specific.
Targets that deploy
These targets publish from CES directly, so end users can query as soon as you deploy.
Snowflake Cortex Analyst Registers a semantic view of the repository with Cortex Analyst for natural-language SQL. Click Preview YAML to read the generated file before you ship it. Deploy expands the card so you can pick the destination database and schema and name the view. The card also shows whether the model validates against Snowflake and when it last checked. See Deploy to Snowflake.
Databricks Publishes the repository as Databricks metric views and a Genie space, so agents query the warehouse natively. Click Build this format, then Deploy. See Deploy to Databricks.
Targets that render formats
These targets produce a file for you to take into another tool. There's no deploy step, because CES doesn't own the destination.
dbt (MetricFlow) Renders the repository as dbt semantic models in MetricFlow format, for you to copy into your dbt project. Click Build this format.
Apache Ossie (incubating) Exports the repository as an Apache Ossie semantic model, a vendor-neutral format you can load into any tool that reads the open standard. Click Build this format, then Get YAML.
When you build, CES checks the file with the validator the Apache Ossie project ships before saving it. A file the validator rejects isn't saved, so what you copy is a file that passes the standard's own checks. CES uses the validator from Ossie commit 89e2e17 (October 27, 2026). The spec is still changing and later versions may reject a file CES exports, so validate against that commit.
Fields with no equivalent in Ossie core, such as sample values, certificate status, and verified SQL questions, travel inside an extension block. The block is marked with the vendor name ATLAN, so a reader that only understands the core format ignores it and everything else still loads. Entries CES can't express at all, such as metrics with filters, aren't silently dropped. They're listed as comments at the top of the file with the reason for each, and the review modal shows a count of them with the YAML. For where each part of the repository lands in the file, see Apache Ossie field map.
Available over MCP
CES doesn't deploy or host an agent. It exposes the context repository over the Atlan MCP server, and any MCP-capable agent reads it at query time.
Claude, OpenAI, Gemini CLI, and other MCP clients Point the agent at the Atlan MCP server, and it reads the repository directly. There's no deploy step, and CES doesn't create or manage the agent.
What to check before you ship
Whatever the target, review the generated format first. An engine can't always represent everything in your model, and CES reports what it left out rather than dropping it silently. See Metrics that aren't emitted.
Apache Ossie field map
CES writes one Ossie file per context repository. The export uses no AI and never queries your warehouse, so the same repository always produces the same file.
| In the context repository | In the Ossie file |
|---|---|
| Repository name, description, and instructions | The model's name, description, and ai_context |
| Tables and views | datasets[] |
| Columns: dimensions, time dimensions, and measures | fields[] on each dataset |
| Relationships | relationships[] |
| Metrics | metrics[] |
| Verified queries | The model's ai_context.examples |
| Row-level filters | Boolean fields labeled filter |
| Anything Ossie core has no field for, such as certificate status, sample values, and the SQL of verified queries | The ATLAN extension |
| Anything CES can't write | A comment at the top of the file, with the reason |
See also
- Create your context repository: build the model these targets are generated from.
- Deploy to Snowflake: the full Cortex Analyst walkthrough, including grants.
- Deploy to Databricks: the full Genie walkthrough, including grants.
- Apache Ossie (incubating): the spec and schema at the version CES validates against.