RouterGrowth MCP¶

Remote MCP server (stateless Streamable HTTP, OAuth or API key) - RouterGrowth puts go-to-market capabilities into any agent as native tools on one balance: SERP, backlink and keyword data, B2B contact and company enrichment, social scraping, ads and outbound. Nothing to install, one key, an exact quote before the first paid run of any call.

Server type: Remote (Streamable HTTP, stateless - POST in, JSON out)
Auth: OAuth 2.1 (PKCE, dynamic client registration, scope "mcp") or an rg_live_ API key
Endpoint: https://api.routergrowth.com/mcp
Tools: discover, inspect, run, batch_run, runs, get_run, history, watch, watches, events, feedback, balance (+ tables and table_write when enabled)
Billing: pay-per-call from the workspace balance; $1 of credit once your email is verified
Rate limit: 300 requests per minute per key, 3,000 per minute per IP

Why This Matters for Operators¶

Growth and sales data normally arrives as four vendor accounts and a pile of wrapper code: an SEO suite for SERP, keywords and backlinks, a contact database for enrichment, a scraper for social, an outbound tool for sequences. RouterGrowth collapses that stack into one MCP surface - a discover, inspect, run loop where the agent finds the capability by job description, reads the exact price and billing conditions, then runs it inside a stated cost cap.

The spending model is the notable part for agents: the server's initialize response instructs the model to inspect before the first run of a call, set max_cost on every item and max_total_cost on a batch, stop and ask before a batch over about $1 unless the user asked for that volume, and never present simulated (rg_test_) data as real. Paid runs are idempotency-keyed, so a retry does not get charged twice.

Tools & Capabilities¶

Group	Tools	What they do
Loop	`discover`, `inspect`, `run`, `batch_run`	Search the catalog by job description; read schema, providers, exact price and billing conditions for one capability; execute a capability with `max_cost`, routing and `idempotency_key`; run one capability for 1 to 200 inputs concurrently with per-item receipts and a total cost cap
Tracking	`runs`, `get_run`, `history`	Run history with attempts, charges and routing reasons; one run by id with long-polling; everything already done to one subject (email, domain, LinkedIn URL, handle, phone, name) with results to reuse
Change watch	`watch`, `watches`, `events`, `feedback`	Save a query and refresh it to get back only new and changed records plus disappeared keys; list watches with last changes; new replies, bounces and messages across connected channels; report a bug or missing capability from the call site
Workspace	`tables`, `table_write`, `balance`	List, describe and query tables by status, due date and data fields; upsert and update rows with `if_status` claims; read wallet balance, reserved and available
Catalog routes	`seo.serp`, `seo.keywords`, `seo.ranked_keywords`, `seo.backlinks`, `contact.find`, `contact.verify`, `company.enrich`, `people.search`, plus social, ads, creative and outbound routes	No DataForSEO, Semrush or Ahrefs account of your own - the agent inspects the price and runs the call

Installation¶

Claude Code (OAuth on first use, or a key header):

claude mcp add --transport http routergrowth https://api.routergrowth.com/mcp

Chat apps (claude.ai, ChatGPT developer mode, Claude Desktop, mobile) add it as a custom connector with the same URL; API-key clients send Authorization: Bearer rg_live_.... For Hermes Agent, add the server to the profile's config.yaml under mcp_servers with the URL and auth: oauth. Full setup pages for every client, including no-code builders and agent frameworks, live on the RouterGrowth quickstart.

Business Relevance¶

One balance, many vendors: SEO data, enrichment and outbound priced per call instead of four subscriptions.
Quote before spend: inspect returns the exact price and billing conditions; max_cost and max_total_cost cap every run.
Batch that behaves: batch_run covers 1 to 200 inputs with per-item receipts and retry-safe idempotency keys.
Change-only refresh: watch returns just the new and changed records since the last run, so recurring monitoring stays cheap.
Agent-safe defaults: the spending rules travel in the handshake, and rg_test_ keys return clearly labeled simulated data.

Integration with CorpusIQ¶

CorpusIQ answers questions about the business's own systems, read-only and cited; RouterGrowth supplies the outside data a growth operator needs next to those answers - the SERP positions, the enriched contacts, the outbound state. An operator can ask CorpusIQ what the site's traffic did and RouterGrowth which keywords moved and which prospects replied, in the same assistant, with each paid call quoted before it runs.

Limitations¶

Paid by design: only discover, inspect, runs, get_run, history, watches, events, tables, feedback and balance are free; run, batch_run and watch spend from the balance.
Stateless server: GET and DELETE answer 405; there is no stream and no session to keep.
tables and table_write appear only when the workspace enables tables.
A test key (rg_test_...) returns simulated data by design - the tools are labeled, never to be presented as real.
Hosted by RouterGrowth and governed by its terms; a live POST initialize returns 401 unauthenticated, confirming it is auth-gated.

FAQ¶

What is RouterGrowth?¶

A pay-per-call API platform that bundles GTM capabilities - SEO (SERP, keywords, backlinks), B2B contact and company enrichment, social scraping, ads and outbound - behind one wallet and exposes them through MCP for agents.

Does the agent need an account with each data vendor?¶

No. RouterGrowth routes the call to its own providers; you hold one balance and the agent quotes the exact price with inspect before the first paid run.

How do cost controls work?¶

Every run takes max_cost; batches take max_total_cost and per-item receipts; the handshake's spending rules tell the agent to stop and ask before a batch over about $1 unless the volume was explicitly requested.

Which clients are supported?¶

claude.ai, Claude Desktop, Claude mobile, ChatGPT (developer mode), Claude Code, Codex, Cursor, VS Code, Gemini CLI, other IDE clients, no-code builders (n8n, Make, Zapier, Copilot Studio, Dify) and major agent frameworks - OAuth where supported, an API key everywhere else.