DataDoe + GitHub CopilotAdd DataDoe to one mcp.json and every developer’s Copilot, plus the cloud agent working your issues, reads Seller Central, Vendor Central and Amazon Ads data. Changes on Amazon stay off until you switch them on.
1{ 2 "servers": { ✓ Running | Stop | Restart3 "datadoe": { 4 "type": "http", 5 "url": 6 "https://mcp.datadoe.com/mcp/v1", 7 "headers": { 8 "datadoe-mcp-key": 9 "${input:datadoe-key}" 10 } 11 } 12 } 13}
.svg)
exports_create — datadoe (MCP Server)exports_get — datadoe (MCP Server)Three SKUs lost money per unit:
| SKU | Price | Net / unit |
|---|---|---|
| LAMP-OAK-L | $34.99 | −$2.10 |
| LAMP-OAK-S | $24.99 | −$0.85 |
| SHADE-LINEN | $12.49 | −$0.31 |
type: AMAZON_LISTINGS_PRICING_UPDATE sku: LAMP-OAK-L · price: 38.49 · dryRun: true
GitHub Copilot reaches MCP servers from the editor, from the cloud agent and from the terminal. Each reads its own config and handles the key differently, so a team rollout touches all three.
A developer asks in the chat panel and watches Copilot pull data, write code and run it.
.vscode/mcp.json, committed to the repoFormerly the coding agent. Assign it an issue and it works in a GitHub Actions runner, then opens a pull request.
COPILOT_MCP_ secret in the copilot environmentThe same agent in a shell, for ops work and quick pulls without opening an editor.
~/.copilot/mcp-config.json or /mcp addJetBrains IDEs, Visual Studio, Xcode and Eclipse take the same server URL and header in Copilot’s MCP settings for that editor.
Create a DataDoe MCP key under Integrations in DataDoe, connect your Amazon accounts, then add the server where your team works. About five minutes for the first developer, a minute or two for the rest.
{ "inputs": [ { "type": "promptString", "id": "datadoe-key", "description": "DataDoe MCP key", "password": true1 } ], "servers": { "datadoe": { "type": "http", "url": "https://mcp.datadoe.com/mcp/v1",2 "headers": { "datadoe-mcp-key": "${input:datadoe-key}"3 } } }}
It holds no secret, so every developer who opens the repo gets the same server. MCP: Add Server in the Command Palette writes the same thing for you.
Seller Central, Vendor Central and Amazon Ads for every brand in your DataDoe organization come through this one endpoint.
VS Code prompts for the DataDoe MCP key the first time the server starts and keeps it out of the file. Start the server, switch the chat to Agent, and pick tools under Configure Tools.
{ "mcpServers": { "datadoe": { "type": "http", "url": "https://mcp.datadoe.com/mcp/v1", "headers": { "datadoe-mcp-key": "$COPILOT_MCP_DATADOE_KEY"1 }, "tools": [2 "sellers_and_vendors_list", "exports_sources_get", "exports_create", "exports_get", "exports_raw_url_get", "sqp_asins_get", "actions_list" ] } }}
Add it as COPILOT_MCP_DATADOE_KEY in the repository’s copilot environment. Only names that start with COPILOT_MCP_ reach the agent, and it does not support OAuth servers, which is why it uses the header.
The cloud agent calls tools without asking anyone. This list leaves out actions_start, so it can read your data and write a dry-run plan into a pull request, but cannot change anything on Amazon.
{ "mcpServers": { "datadoe": { "type": "http", "url": "https://mcp.datadoe.com/mcp/v1", "headers": { "datadoe-mcp-key": "YOUR_DATADOE_MCP_KEY"1 }, "tools": ["*"]2 } }}
This file sits in your home directory, so the key never lands in a repository. Typing /mcp add inside the CLI fills it in for you.
The CLI asks before it runs a tool, so allowing every DataDoe tool here is reasonable for a person at the keyboard. Write Actions still need switching on in DataDoe first.
Building alone rather than for a team? The Cursor guide covers a one-person setup, and the Codex guide covers OpenAI’s coding agent.
DataDoe is a data and action layer: Copilot can change listings, prices, FBM orders and Amazon Ads campaigns through it. That is why each layer below sits with a different person. The dry run is optional: you ask for it in the call.
Each connected Seller, Vendor or Ads account is read only or read and write. Pilot a team on read only and nothing can change.
Every Action type is off until someone switches it on. Price updates are their own type, separate from title and bullet edits.
Copilot can send any change with a dry run, which checks the payload without submitting it to Amazon. Put the rule in your repo instructions (.github/copilot-instructions.md) so it does that by default.
Agent mode and the CLI ask before each tool call. The cloud agent does not, so its tool list leaves the write tool out.
Each change is recorded with its payload and result, and the history comes back through the same server. There is a ceiling of 500 live Actions per organization.
Amazon credentials stay in DataDoe. Copilot only ever holds the DataDoe key, which you can revoke.
On Copilot Business or Enterprise, three people touch the rollout before a developer sees a single DataDoe tool. Here is who does what, in order.
Organization Settings → Copilot → Policies → MCP servers in Copilot. It is off by default on Copilot Business and Enterprise, and until it is on, no seat in the org can load DataDoe.
If MCP access is restricted to registry servers, DataDoe has to be on the registry your admins approve before it will start.
Paste the cloud agent JSON under Repository Settings → Copilot → MCP servers and add COPILOT_MCP_DATADOE_KEY to the copilot environment.
Keep accounts read only for the pilot. Later, switch on only the Action types the team needs, such as Amazon Ads bids before prices.
"datadoe": { "type": "http",
"url": "https://mcp.datadoe.com/mcp/v1", … }Save MCP configurationWrite the issue the way you would brief a developer, assign it to Copilot, and review what comes back. With DataDoe in its config, it checks real Seller Central, Vendor Central and Ads data while it works instead of guessing from a fixture.
Use DataDoe orders, fees, refunds and Sponsored Products spend with our COGS. Add a job under jobs/seller/ that posts the list to #brand-ops at 07:00, with the three biggest drivers for each SKU.
exports_sources_getexports_create to test the mathsCompare Vendor Central purchase orders with Amazon’s demand forecast by ASIN for the last eight weeks. Write a CSV to reports/vendor/ and a three-line summary for the account manager.
Per campaign, list the search terms to add as negative exact. Output the plan as Sponsored Products payloads with dryRun: true. Do not apply anything; we review the plan in the PR.
actions_start to a person in agent modeNorthfield connected Germany and France in DataDoe yesterday. Add both to clients/northfield.yml and make sure the report converts to EUR per marketplace, not to USD.
sellers_and_vendors_listClick an issue to see the brief and the work. Plans that would change Amazon arrive as dry-run payloads for a person to apply.
This is how agencies and in-house brand teams on Copilot turn an Amazon question into code they own. Copilot uses DataDoe to explore and to write the code; the finished job calls the DataDoe REST API or BigQuery on a schedule.
Which SKUs lost margin last week, and what drove it?
Where is Amazon ordering below its own forecast?
Which search terms spend with no orders behind them?
dryRun payloadsWhat changed for Northfield this week?
Steps 1 to 3 use the MCP server. Step 4 is ordinary code your team reviewed; it changes nothing on Amazon unless a person applies a write on purpose, as in the Amazon Ads row.
Your DataDoe key and data work in every MCP client, so people on the team who don’t live in GitHub can use the same Amazon data.
Add it to .vscode/mcp.json in your repository, or run MCP: Add Server from the Command Palette. For DataDoe, use a server of type http with the URL https://mcp.datadoe.com/mcp/v1 and a datadoe-mcp-key header that reads the key from an input. Start the server, switch the chat to Agent, and the DataDoe tools appear under Configure Tools.
In DataDoe, under Integrations, after you connect your Amazon accounts. The key gives access to your DataDoe organization, not to your Amazon login: Amazon credentials stay with DataDoe. You can revoke the key and create a new one at any time.
The MCP servers in Copilot policy is off by default for Copilot Business and Enterprise. An organization owner turns it on under Settings, Copilot, Policies. If MCP access is limited to registry servers, DataDoe also has to be on the registry your admins approve. The policy does not apply to personal Copilot plans.
Yes. GitHub now calls it the Copilot cloud agent. Paste the server JSON under Repository Settings, Copilot, MCP servers, and store the key as COPILOT_MCP_DATADOE_KEY in the copilot environment. The cloud agent calls tools without asking, so give it a list of read tools and leave out actions_start.
Only through DataDoe Actions, and only when you allow it. Every Action type is off by default, the Amazon account has to be set to read and write, and price updates are a separate Action type. In agent mode Copilot asks before each tool call, any change can be previewed with a dry run first, and every Action is logged.
Yes. DataDoe puts Seller Central, Vendor Central and Amazon Ads in one schema with 130+ tables, including purchase orders, Retail Analytics, Search Query Performance and campaign data, so one server covers all three.
One DataDoe plan at $97 a month after a 14-day free trial, with unlimited Seller, Vendor and Ads connections and 10 million data rows included. Extra rows are added automatically at $49 per 10 million, and extra seats are $29 a month each. Copilot is billed separately by GitHub.
We'll set up the config with your admin, connect a client account and show the cloud agent working an Amazon issue.
Book a demoOne server for agent mode, the cloud agent and the CLI. Seller, Vendor and Ads data for the whole team — and writes only where you allow them.
We use cookies to improve your experience and analyze traffic. By clicking "Accept", you agree to our use of cookies. Read the Cookie Policy.