SignalKit Connector for Claude

Ask Claude about your brand's mention rate, which sources the platforms cite, what they searched for, what is open on the alert shelf, how your site audits scored and what is in the content queue; add prompts, close alerts, start an audit and write a draft — all from inside the chat. Any connection can also discover the registered app operations and run the ones its credential may reach. Signed in with OAuth, an agent can do ordinary work your role allows. Clinical certification and publishing require you to act in the app; changing billing also needs the owner's permission there.

Install (Claude.ai + Claude Desktop)

1. Open Claude → Customize → Connectors → + → Add custom connector.
2. Paste this URL:

   https://app.signalkit.ai/api/mcp

3. Click Add. Claude will redirect you to sign in to SignalKit (via Clerk). Approve, and you'll bounce back to Claude with the connector installed.
4. Try it: ask Claude "How is my brand doing in AI search this week?"

Set up any MCP client

Clients that take a remote server by URL connect to the hosted endpoint directly and sign in with OAuth. For a client that cannot sign in, or a narrower credential, create a key at account / API keys — a read key for an agent that reports, a write key only when it should add prompts, close alerts or start audits — and use the API-key variant for your client, which reads the key from SIGNALKIT_API_KEY or prompts for it.

claude mcp add --transport http signalkit https://app.signalkit.ai/api/mcp

claude mcp add --transport http signalkit https://app.signalkit.ai/api/mcp \
  --header "Authorization: Bearer $SIGNALKIT_API_KEY"

{
  "mcpServers": {
    "signalkit": {
      "url": "https://app.signalkit.ai/api/mcp"
    }
  }
}

{
  "mcpServers": {
    "signalkit": {
      "url": "https://app.signalkit.ai/api/mcp",
      "headers": { "Authorization": "Bearer ${env:SIGNALKIT_API_KEY}" }
    }
  }
}

{
  "servers": {
    "signalkit": {
      "type": "http",
      "url": "https://app.signalkit.ai/api/mcp"
    }
  }
}

{
  "inputs": [
    {
      "type": "promptString",
      "id": "signalkit-key",
      "description": "SignalKit API key",
      "password": true
    }
  ],
  "servers": {
    "signalkit": {
      "type": "http",
      "url": "https://app.signalkit.ai/api/mcp",
      "headers": { "Authorization": "Bearer ${input:signalkit-key}" }
    }
  }
}

codex mcp add signalkit --url https://app.signalkit.ai/api/mcp
codex mcp login signalkit

codex mcp add signalkit --url https://app.signalkit.ai/api/mcp \
  --bearer-token-env-var SIGNALKIT_API_KEY

Keep the key out of anything you commit. A key can carry an expiry; set one for an agent you will not be watching. Then try: "Which sources cite my competitors but not me?"

Local stdio, from a source checkout

packages/mcp in the source tree is a stdio server over the same REST API, for a client that runs a command rather than taking a URL. It is not published to npm. Build it (pnpm install --frozen-lockfile && pnpm build in that directory) and point the client at the built file:

{
  "mcpServers": {
    "signalkit": {
      "command": "node",
      "args": ["/absolute/path/to/packages/mcp/dist/index.js"],
      "env": { "SIGNALKIT_API_KEY": "sk_live_…" }
    }
  }
}

Auth

SignalKit's MCP server speaks OAuth 2.0 with Dynamic Client Registration (RFC 7591). The discovery endpoint is:

https://signalkit.ai/.well-known/oauth-authorization-server

An OAuth token is the person's own credential: every tool, with exactly the projects and role that person has in the organization, including drafts, brands, members and API keys. Clinical certification and publishing require a signed-in human action in the app. Billing changes run only if the owner has turned on “Allow agents to change billing” under Account; no agent can turn that on. Over an API key, that work returns a dashboard link instead. An API key is the alternative for a client that cannot sign in, and it can be narrower on two independent axes, chosen when it is minted and not editable afterwards:

• Scope — which projects it reaches: every project in the organization, or a fixed set.
• Access mode — what it may do inside them. A read key is offered only the read tools in the table below (tools/list omits the rest) and is refused a write tool if it calls one anyway. A write key adds the tools that create, change, run, close or spend. New keys are read-only unless they ask for write.

Neither axis widens the other, and a key with an expiry is refused before any tool runs once it has passed. The key goes in the header:

Authorization: Bearer sk_live_…

Tools

The connector exposes 57 tools: 33 reads and 24 writes. If your OAuth token has no organization claim, call list_organizations and pass its id as organizationId to tenant tools. Project tools use your default project unless you pass a projectId arg; run_site_audit requires one, because the audit and its spend are booked to it. Reads never trigger paid work: get_opportunities serves the stored brief and get_health the stored check outcomes.

Start with list_app_operations before using read_app or change_app. Every operation lists the parameters its route reads; a route on the operation registry also publishes the validation schema it enforces, and for the others the route stays authoritative for types, required fields and nested objects. An operation your connection may not perform returns the relevant dashboard URL without making the change.

Some clients cache tool schemas for a conversation. If a newly added parameter is rejected before SignalKit receives the call, reconnect the connector or start a new conversation to load the current schema.

Programmatic discovery

The connector advertises a standard MCP server card at /.well-known/mcp/server-card.json with tool names, OAuth endpoints, and capability flags. Anthropic's marketplace reads this card; other MCP clients can too.

Example session (raw JSON-RPC)

For curl-driven testing, the streamable-HTTP endpoint takes a JSON-RPC request and returns a JSON-RPC response:

# Initialize
curl -sX POST https://app.signalkit.ai/api/mcp \
  -H "Authorization: Bearer sk_live_…" \
  -H "Content-Type: application/json" \
  -d '{"jsonrpc":"2.0","id":1,"method":"initialize","params":{}}'

# List tools
curl -sX POST https://app.signalkit.ai/api/mcp \
  -H "Authorization: Bearer sk_live_…" \
  -H "Content-Type: application/json" \
  -d '{"jsonrpc":"2.0","id":2,"method":"tools/list"}'

# Call a tool
curl -sX POST https://app.signalkit.ai/api/mcp \
  -H "Authorization: Bearer sk_live_…" \
  -H "Content-Type: application/json" \
  -d '{"jsonrpc":"2.0","id":3,"method":"tools/call","params":{"name":"get_overview","arguments":{}}}'

# Start a paid audit — a retry with the same idempotencyKey buys nothing
curl -sX POST https://app.signalkit.ai/api/mcp \
  -H "Authorization: Bearer sk_live_…" \
  -H "Content-Type: application/json" \
  -d '{"jsonrpc":"2.0","id":4,"method":"tools/call","params":{"name":"run_site_audit","arguments":{"projectId":"<uuid>","url":"https://example.com","idempotencyKey":"audit-2026-09-16"}}}'

Support

Bug reports + feature requests: hello@signalkit.ai.