Model Context Protocol

Customer research, inside your agent.

Connect Claude, Cursor, or Windsurf directly to Sovrn AI.

Browser sign-in for interactive clients. No key to paste. Grounded in a model representing ~759M people.

Connect

Add one URL. Sign in in your browser.

Choose your client for the exact setup. OAuth is the default; API keys appear only on the HTTP path for scripts and CI.

1

Add the server

terminal
claude mcp add --transport http sovrn-lbm https://sovrn.ai/api/mcp

Run once, restart Claude Code, then type /mcp and pick Authenticate. Your browser opens the Sovrn sign-in.

2

Sign in when prompted

No key needed. Your browser opens Sovrn sign-in, then asks you to approve tool access. Audience tools are read-only; if you've connected Google Ads in Settings → Connectors, approval also lets this client preview and, after you confirm, apply changes to it.

3

Ask for an audience

  • “Find sneaker buyers who follow streetwear releases and compare sizing charts.”
  • “Who overlaps between my ski buyers and luxury travelers?”

You describe the people. Your agent chooses the tool.

Live tool catalog

Twenty tools. One audience model.

This list mirrors the server's registered tools. The exact list a client receives can be smaller when a role or plan does not include a tool.

Start here

Find and understand your customers.

search_audience

Describe who and what you want in plain English. The server returns the match, a fit score, behavioral evidence, and related groups — no headcount. Follow-up tools compare, refine, or inspect the result.

estimate_audience_sizeFind

estimate_audience_size (unavailable)

Not available on the current index — every call is refused. Use fit (summary.avgScore) and depth (summary.concentration.depth) from search_audience instead; there is no population count.

discover_cohortsFind

Explore related groups

Test themes against the people you found and see which groups genuinely over-index.

audience_time_seriesCompare

See change over time

Compare interest by hour or day with coverage-normalized periods.

compute_overlapCompare

Compare two customer groups

Measure the observed overlap and its share of each group.

search_audience_excludeCompare

Remove a group

Find a group, subtract people you do not want, and keep the cleaner result.

sample_usersInspect

Quality-check a match

Inspect a small sample from a result to verify the behavioral evidence behind it.

describe_userInspect

Inspect one result

Read the behavioral metadata for one pseudonymous result from an earlier search.

list_customer_profilesProfiles

List saved Customer Profiles

See the Customer Profiles saved in your workspace. Read-only — building or editing a Profile stays in the dashboard.

get_customer_profileProfiles

Open a Customer Profile

Get one Profile's spec, its latest run, and its week-over-week trends. Deliberately no headcount on this output — no matched-users count, no size, no fit score; read shares and trend direction instead.

generate_audience_briefProfiles

Write a grounded brief

Generate the same server-side brief the dashboard writes for a search — who they are, where to reach them, what works, watch-outs. Durable: a cold call may return pending; free to poll.

get_capabilitiesLearn

Read the server guide

Load the data model, limitations, metric definitions, and additional-skill manifest.

read_skill_docLearn

Open a skill document

Fetch a Sovrn reference document when the client cannot read MCP resources directly.

Census context, where present, describes areas in aggregate, not individual traits. There is no population count or market size — results are reported as fit and depth. Small groups are directional.

For scripts and builders

Use the protocol directly when you need to.

Plain HTTP

JSON-RPC 2.0 over POST. This secondary path uses an API key because scripts cannot complete browser sign-in.

API keys
curl
curl -X POST https://sovrn.ai/api/mcp -H "Authorization: Bearer YOUR_KEY" -H "Content-Type: application/json" -d '{"jsonrpc":"2.0","id":1,"method":"tools/call","params":{"name":"search_audience","arguments":{"prompt":"people shopping for trail-running gear and reading endurance training"}}}'

Built to be discovered

A client can learn the server without a separate SDK or hardcoded integration guide.

tools/list
Typed schemas filtered to the account’s permissions and plan.
resources/list
Skill and data-model documents the agent can read when needed.
pending → retry
Long searches continue server-side; retry the call to collect the result.
Ready to connect

Bring your customer questions into the tool you already use.

Add the server, sign in in your browser, and ask in plain English.

Connect your client