Use Sidecar MCP in an AI client

The Sidecar MCP server gives AI agents live access to your data catalog: dbt models, warehouse tables, dashboards, column-level lineage, and your team’s knowledge base. Ask “what feeds the revenue dashboard?” or “what breaks if I change this column?” and the agent answers from the same cross-tool context Sidecar has.

This is not the Agent Console MCP setup. To give Sidecar agents access to an external MCP server, see Connect an MCP server.

Server URL: https://service-mcp.sidecardata.com/sidecar/mcp

Connect with OAuth (recommended)

No token needed — you’ll be prompted to log in with your Sidecar account in the browser.

Claude Code
Claude (web)
Cursor
VS Code

Run the following command in your terminal:

claude mcp add --transport http sidecar-mcp https://service-mcp.sidecardata.com/sidecar/mcp

Then run /mcp inside a Claude Code session and follow the login prompt.

In claude.ai, go to Settings → Connectors.
Click Add custom connector.
Paste https://service-mcp.sidecardata.com/sidecar/mcp and click Add.
Click Connect and log in with your Sidecar account.

Open Cursor Settings → MCP and add a new server with URL https://service-mcp.sidecardata.com/sidecar/mcp.
Cursor detects OAuth support — click Login and approve access in the browser.

Run the following command in your terminal:

code --add-mcp '{"name":"sidecar-mcp","type":"http","url":"https://service-mcp.sidecardata.com/sidecar/mcp"}'

Start the server from the Extensions view or when prompted in chat; VS Code walks you through the OAuth login.

Connect with a token

Use a token for clients without OAuth support, service accounts, or CI.

1. Create a token

In the Sidecar app, navigate to the Users tab.
Click the Tokens tab.
Click Create Token.
Enter a name for the token.
Set the expiration (1–365 days, or leave empty for no expiration).
Click Create.

The token is only visible once. Copy it and store it in a safe place.

2. Configure your client

Codex
Claude Code
Other AI IDEs

Add the following to ~/.codex/config.toml. The Codex CLI and IDE extension share this file:

[mcp_servers.sidecar]
url = "https://service-mcp.sidecardata.com/sidecar/mcp"
bearer_token_env_var = "SIDECAR_MCP_TOKEN"

Then set SIDECAR_MCP_TOKEN in your shell environment to the token you created in Step 1.

Run the following command in your terminal:

claude mcp add --transport http sidecar-mcp https://service-mcp.sidecardata.com/sidecar/mcp \
  --header "Authorization: Bearer scp_mcp_rest_of_generated_token"

Replace scp_mcp_rest_of_generated_token with the token you created in Step 1.

Use the following details to manually configure the MCP server in your IDE:

Field	Value
URL	`https://service-mcp.sidecardata.com/sidecar/mcp`
Authorization	`Bearer scp_mcp_rest_of_generated_token`

Replace scp_mcp_rest_of_generated_token with the token you created in Step 1.

Teach your agent when to use Sidecar

Connecting the server makes Sidecar’s tools available, but your agent might not use them unless you tell it when they matter. Add the following block to the CLAUDE.md or AGENTS.md file in your data project. In a monorepo, add it to the instruction file closest to your dbt or analytics code.

## Use Sidecar for data context

Use the connected Sidecar MCP proactively whenever a task involves warehouse
data, dbt models or sources, BI assets, lineage, or business definitions.

- When a table, model, source, column, dashboard, metric, or business concept is
  named, resolve or search for it in Sidecar. Ground your answer in catalog
  metadata instead of guessing identifiers, schemas, owners, or definitions.
- Before planning a refactor or changing warehouse SQL, a dbt model, a source
  definition, or a schema contract, inspect the relevant upstream and downstream
  lineage. Inspect column-level lineage when a column changes.
- After making one of those changes, calculate downstream impact. Report the
  affected models, tables, dashboards, and other consumers before declaring the
  work complete.
- For open-ended discovery, start with the catalog overview, then search and
  fetch detailed asset context as needed. Use linked tests, tickets, team
  discussions, knowledge-base content, and Context Garage content when relevant
  to explain business intent.
- If Sidecar returns multiple matches, disambiguate them using their identifiers
  and source systems. Do not silently choose a similarly named asset.
- Treat Sidecar as the source of truth for cataloged metadata and cross-tool
  lineage. Treat repository files as the source of truth for unsynced local
  code, and call out when the two might differ. Sidecar is read-only; use other
  authorized tools for live data queries or changes.
- Summarize the Sidecar evidence that informed the work, including the resolved
  asset identifiers and any important downstream impact. If Sidecar is
  unavailable or finds no match, say so and continue with the best local
  evidence.

If your AI client supports reusable skills or rules, you can use the same text there instead. Scope it to data and dbt work so Sidecar is invoked when its catalog context is useful.

What you can ask

Once connected, try prompts like:

“What data do we have about customers?”
“What feeds the revenue dashboard, and when did its sources last refresh?”
“If I change orders.customer_id, what breaks downstream?”

Available tools

Tool	What it does
`get_catalog_overview`	Summary of the catalog: connected tools, asset counts, tags, company context
`resolve_asset_id`	Look up an asset by name — schema, columns, description
`advanced_search`	Find assets by identifier, keyword, or semantic similarity
`describe_asset`	Deep-dive on one asset: overview, columns, tests, tickets, Slack context
`inspect_lineage_details`	Upstream and downstream lineage graph for an asset
`calculate_downstream_impact`	Blast-radius analysis for a change
`get_column_lineage`	Column-level lineage for a specific column
`get_context_repo`	The semantic contract deployed to this token’s Context Garage
`list_knowledge_base_files` / `get_knowledge_base_file_content`	Browse and read your team’s knowledge base
`load_skill`	Load a Sidecar development guide into the agent’s context

All tools are read-only — the server never modifies your catalog or warehouse.

Troubleshooting

401 / authentication errors — your token expired or the OAuth session lapsed. Re-run the login (OAuth) or create a fresh token.
Search returns nothing — call get_catalog_overview first to see what’s actually in the catalog for your account; your user may not have access to the workspace you expect.
OAuth login never opens or loops — confirm the URL is exactly https://service-mcp.sidecardata.com/sidecar/mcp and that your client is up to date. Codex does not currently support OAuth with Sidecar — use a token instead.
Still stuck? Contact support@sidecardata.com.

Sidecar is also listed in the official MCP Registry as com.sidecardata/sidecar. Clients that browse the registry can install it from there.

Source control

Data Warehouses

Transformation

BI & Analytics

Extract & Load

Communication

MCP

Project Management

Use Sidecar MCP in an AI client

Connect with OAuth (recommended)

Connect with a token

1. Create a token

2. Configure your client

Teach your agent when to use Sidecar

What you can ask

Available tools

Troubleshooting

​Connect with OAuth (recommended)

​Connect with a token

​1. Create a token

​2. Configure your client

​Teach your agent when to use Sidecar

​What you can ask

​Available tools

​Troubleshooting

Connect with OAuth (recommended)

Connect with a token

1. Create a token

2. Configure your client

Teach your agent when to use Sidecar

What you can ask

Available tools

Troubleshooting