MCP connector
Connect your workspace to an MCP-capable AI client so it can read your citations, competitors, changes and forecasts, and act on them.
MCP is the Model Context Protocol: a standard way for an AI client to call tools. Outercite runs an MCP server, so a client such as Claude or Cursor can work with your workspace directly instead of you copying numbers into a chat.
What you'll learn
- What the connector exposes
- How to connect it and how authentication works
- What it is good for, and what it is not
What it exposes
Tools in five groups:
| Group | Examples of what it can do |
|---|---|
| Reading your position | Dashboard stats, citation lists and trends, platform coverage |
| Evidence | Individual citations, the sources behind them, the latest brief |
| Competition | List competitors, add one |
| The loop | List changes, read one, list actions, mark an action complete, trigger a check run |
| Plumbing | List, create and delete webhooks |
Agencies get a portfolio group as well: list clients, compare them, portfolio health, recent activity, and generating a client brief.
The exact tool list is discoverable from the client, which is the right place to read it: it is the one list that cannot go out of date.
Connect it
The server lives at /api/mcp on the app origin. Authentication is either your API key as a bearer token, or OAuth if
your client supports it. A 401 advertises where the OAuth metadata lives, so a compliant client can discover it and
complete the flow on its own.
Two things follow from the key being an ordinary API key:
- Permissions apply. A read-only key gets read-only tools. If you want the client to add prompts, the key needs write permission.
- Tenancy applies. The connector sees the workspace the key belongs to, and nothing else.
The connector is available on paid plans. See plans and limits.
What it is good for
- Asking questions in the client you already work in, with your real data underneath rather than a paste.
- Bulk work. "Add these twelve prompts to the topic Pricing" is a sentence, not an afternoon.
- Drafting with evidence in context. Pulling the sentences an engine actually wrote into whatever you are writing.
What it is not
- It is not the assistant. The in-app assistant grounds every answer in counted rows and opens the evidence beside it. An MCP client is a general model calling tools, so the discipline of the answer is the client's, not ours.
- It is not a bypass of your plan. The same gates, limits and rate limits apply.
- It is not WebMCP. That is a different thing: public tools for in-browser agents on our marketing site, with no account and no auth. See WebMCP.
Try this in Outercite
Create a read-only key in API settings, connect it in your MCP client, and ask it for your platform coverage. If the numbers match what the assistant tells you, the connection is scoped correctly.
Related
API access
Keys and permissions, which the connector reuses.
WebMCP
The other MCP in this product, and why it is unrelated.
Meet the assistant
The grounded alternative inside the product.
Last updated on
API access
Create a key, authenticate with a bearer token, and pull citations, prompts, competitors, changes, forecasts and Trace data into your own tools.
WebMCP
The browser standard that lets a web page hand typed, callable tools to an in-browser AI agent. What it is, where it actually stands, and whether you should implement it yet.