Agent Context MCP Public Preview
Fivetran Agent Context MCP exposes your context layer to AI agents, coding assistants, and AG-UI applications through the Model Context Protocol (MCP).
Overview
Fivetran Agent Context MCP is hosted at https://api.fivetran.ai/mcp.
Agent Context MCP is a governed interface that uses the MCP to expose the context layer you created with Fivetran Context Layer to AI tools. It enables any AI client to retrieve answers based on your business definitions, metrics, and operational data. Instead of granting every AI tool direct access to your data warehouse, connect them all to a single secure interface, the Agent Context MCP.
Agent Context MCP supports AI coding and chat tools, such as Claude, Claude Code, Cursor, ChatGPT, and Codex. It also supports AG-UI applications, such as Slack bots, Zendesk apps, and internal web assistants.
Model Context Protocol (MCP)
Model Context Protocol (MCP) is an open-source standard that gives AI applications a consistent way to connect to external systems. Learn more in Anthropic's What is the Model Context Protocol (MCP)? documentation.
Fivetran Agent Context MCP is a server that implements the MCP to give AI applications access to your organization’s context layer.
How it works
Once you've built your context layer with Fivetran Context Layer, do the following to use Agent Context MCP:
- Connect your AI client or app: Connect Agent Context MCP to your organization's AI coding or chat tools, or to an AG-UI application.
- Query your context layer: Ask questions in your connected AI tool or app. Agent Context MCP retrieves grounded answers from your context layer, such as business definitions, metrics, and operational data.
For detailed steps, see the Agent Context MCP Setup Guide.
Available tools
Agent Context MCP exposes the following tools to your connected AI client:
| Tool | Type | Description |
|---|---|---|
_instructions | Read | Required. Returns the system prompt with tool usage instructions and response format. Your AI client calls this before any other tool. |
| Source search tools | Read | Search a connected documentation source or system of record, such as Salesforce, Notion, Zendesk, or Jira, for relevant results. Agent Context MCP exposes one search tool per connected source. |
search_tables | Read | Search catalog tables by business concept, metric, or analytics intent. |
search_concepts | Read | Search catalog concepts by business vocabulary, stage, or metric terms. |
list_tables | Read | List catalog tables in the accessible connection scope, optionally narrowed to a schema. |
list_schemas | Read | List schemas available in the accessible connection scope. |
organizational_fingerprint | Read | Read the organization fingerprint for questions about your organization or company. |
schema_instructions | Read | Fetch optional human-authored instructions for a catalog schema. |
table_deep_dive | Read | Inspect a single catalog table, including its metadata and columns. |
execute_aisql | Read | Execute a read-only AISQL query against the catalog index. |
provide_feedback | Write | Submit feedback about a response, such as when an answer is incorrect or unsatisfactory. |
This tool interface is subject to change as AI models evolve, to ensure we give your AI clients more accurate and efficient access to your context layer.
AISQL
AISQL extends standard SQL with AI functions that bring semantic search directly into your queries:
semantic_relevance(column, 'query'): Ranks rows by semantic similarity to a natural language query.
AISQL queries stay read-only.
Prerequisites
To use Agent Context MCP, you need:
- A Context Layer built for your organization.
- Access to Agent Context MCP, granted by your Fivetran account admin.
Additional support
If you have questions about Agent Context MCP or need help connecting it to your AI tools, contact Fivetran Support.