Skip to main content
Adriel MCP connects AI assistants such as Claude and ChatGPT to the marketing data in an Adriel workspace. Ask a question in natural language, for example “compare spend and conversions by channel for last month”, and the AI assistant pulls the numbers from the data sources connected in Adriel and answers in the conversation. The Model Context Protocol (MCP) is an open standard that lets AI assistants call external tools and data. Adriel MCP is a read-only MCP server: it fetches data and context from Adriel and never changes anything in the workspace. To set it up, see Connect Adriel MCP.

​
When to use Adriel MCP

Adriel MCP fits teams that already work in an AI assistant and want Adriel data available there without exporting CSVs or copying numbers from a dashboard. Common uses include:
  • Pulling cross-channel performance into a proposal, client report, or slide outline.
  • Drilling from a monthly overview down to weekly or daily numbers for one channel.
  • Combining Adriel data with other files or tools already connected to the same AI assistant.
For questions asked inside Adriel, with a dashboard or widget attached as context, use the AI Agent instead.

​
Supported AI assistants

Adriel MCP currently supports:
  • Claude (claude.ai, Claude Desktop, and Claude Code)
  • ChatGPT
See Connect Adriel MCP for setup steps.

​
Before you start

Adriel MCP only returns data that already exists in Adriel and that the signed-in member is allowed to see. Before connecting an AI assistant:
  • Every person using Adriel MCP needs their own Adriel login. The AI assistant signs in to Adriel with OAuth, using the member’s own account. There is no shared key or service account.
  • Connect the data sources to query. Adriel MCP reads from data sources already connected in the workspace. See Data sources overview.
  • Assign data sources and members to teamspaces. Adriel MCP uses the existing teamspace structure to decide what each member can query. See the next section.

​
How access works

Each Adriel MCP connection is tied to one Adriel member: the account that signed in. Adriel MCP is available on all Adriel plans, and every workspace role (View only, View and edit, and Admin) can use it to read data. Inside the workspace, Adriel MCP follows the same access rules as dashboards:
  • A member can only query teamspaces they belong to.
  • Inside a teamspace, a member can only query the data sources assigned to that teamspace.
  • If a member is excluded from a data source, that data source is not available to them through Adriel MCP either.
  • Admin members can access every teamspace in the workspace.
This makes teamspaces the guardrail for what an AI assistant can pull. For example, an agency can put each client’s ad accounts in a separate teamspace and invite only that client’s team to it. A member of the Client A teamspace can’t pull Client B data through Adriel MCP, even though both clients’ accounts are connected to the same workspace.
Before rolling Adriel MCP out to a team, review each teamspace’s Data sources and Members in Teamspace settings. Whatever a member can see in a teamspace, they can query from an AI assistant.

​
What Adriel MCP can do

  • List teamspaces and data sources the member has access to.
  • Read teamspace context, such as teamspace notes and brand profiles, so the AI assistant can use the same business context the team uses in Adriel.
  • Show available metrics and breakdowns for the selected data sources, including the workspace’s custom metrics.
  • Query aggregated performance data, such as spend, impressions, clicks, and conversions, by date (day, week, or month), channel, and other breakdowns. One request can return several named datasets.
  • Analyze returned datasets with SQL: filter, sort, calculate, aggregate, and join completed datasets without fetching the data again.
  • Report what was queried. Each result includes the data sources, metrics, breakdowns, filters, date range, time zone, and currency that were used, so the numbers can be checked.
Adriel MCP reads the same data that dashboards show, so numbers match a dashboard with the same data sources, date range, metrics, and filters. Ad performance data is already normalized across platforms in Adriel. Shared metrics such as spend, impressions, and clicks mean the same thing for every ad platform, so cross-channel comparisons work without mapping fields by hand.

​
What Adriel MCP can’t do

  • Change anything in Adriel. Adriel MCP can’t create or edit dashboards, widgets, data sources, teamspaces, custom metrics, or alerts, and it can’t change campaigns on the ad platforms.
  • Access data outside the member’s teamspaces, or data sources that aren’t connected in Adriel.
  • Return metric formulas. Results show metric names, not how each metric is calculated. Check custom metric definitions in Adriel.

​
How AI assistants use Adriel MCP

The AI assistant decides which tools to call and in what order. Describe the outcome in natural language and the assistant runs the tools. A typical sequence looks like this:
1

Find the teamspace

The assistant calls list_teamspaces to see which teamspaces the member can access, then get_teamspace_context to read that teamspace’s notes and brand profiles.
2

Find data sources and fields

The assistant calls list_data_sources to see the data sources in the teamspace, and list_fields_and_breakdowns to see which metrics and breakdowns are available.
3

Query the data

The assistant calls query_data with a description of the data needed. Larger queries may take a moment; the assistant checks progress with get_query_result.
4

Analyze the results

The assistant calls query_dataset_sql to filter, sort, or combine the returned datasets before answering.

​
Available tools

All tools are read-only.

​
Tips for reliable results

The more specific the request, the more consistent the answer. A request that names the following four things is run exactly as written:
  • Data sources — name them as they appear in Adriel, for example “Google Ads and Meta Ads”.
  • Date range — use exact dates, for example “2026-10-01 to 2026-10-31”.
  • Metrics — for example “spend, clicks, and conversions”.
  • Granularity — for example “by month and by channel” or “by day”.
Example prompts:
After a query runs, ask the assistant to show which data sources, metrics, breakdowns, and filters it used. If a number looks wrong, that’s the first thing to check.

​
Limits

  • Result size. Each response returns up to 500 rows and 70 KB. For larger results, ask for a coarser breakdown (for example, weekly instead of daily), fewer breakdowns, or a shorter date range.
  • Result retention. Query results and their download links are available for 15 minutes. After that, the assistant runs the query again, and the numbers reflect the latest data in Adriel.

​
Limitations

Adriel MCP is new, and AI assistants sometimes fill in parts of a request that weren’t specified. Known behaviors:
  • Unrequested filters. When a request is vague, a filter that wasn’t asked for can be added, for example matching campaign names against a data source name. This can return an empty result.
  • Similar metrics swapped. A related metric can be picked instead of the one intended, for example All conversions instead of Conversions.
  • Channel granularity varies. Grouping “by channel” can return one row per platform (for example, Google) or one row per campaign type (for example, Google Search and Google Video).
  • Data source IDs. Referring to a data source by its numeric ID can fail to match. Refer to data sources by name.
Naming the data sources, date range, metrics, and granularity in the request avoids most of these.

​
Troubleshooting

Adriel MCP only shows teamspaces the member belongs to and data sources assigned to those teamspaces. Check the teamspace’s Data sources and Members in Teamspace settings, and check whether the member is excluded from the data source in Permissions and access. Also confirm the AI assistant signed in with the right Adriel account.
Ask the assistant to show the filters it used. If it added a filter that wasn’t requested, ask it to run the query again with no filters. Also confirm the data source has data for the selected date range.
The request likely left the metric, breakdown, or filter open, and it was filled in differently each time. Name the data sources, date range, metrics, and granularity explicitly, and compare the conditions each run used.
Compare the date range, time zone, currency, metrics, and filters the query used with the dashboard’s settings. See Numbers don’t match the source platform for other common causes.

Connect Adriel MCP

Add Adriel MCP to Claude or ChatGPT.

Teamspace settings

Control which members and data sources each teamspace includes.

Permissions and access

Exclude specific members from a data source.

AI Agent

Ask questions about connected data inside Adriel.