Quickstart: agents (SKILL.md)

Add lead enrichment, SEO, social and outbound tools to Claude Code, Codex, Cursor or any agent with one SKILL.md (Agent Skills) instead of writing API wrappers: discover, inspect and run, with the spending rules built in.

RouterGrowth is built to be called by agents. The integration is one file: SKILL.md teaches an agent the MCP tools, the CLI, the HTTP interface, the spending rules (inspect before run, cap with max_cost, ask before batches over about $1) and which workflow skill to load for a whole GTM job.

One-line install

Tell your agent:

Set up https://www.routergrowth.com/SKILL.md

The skill contains its own install instructions, so the agent persists it in the right place for its harness:

Then connect the agent to your workspace. Two ways:

• OAuth through the MCP server, no key to handle. Claude Code: claude mcp add --transport http routergrowth https://api.routergrowth.com/mcp, then /mcp and Authenticate. Codex: codex mcp add routergrowth --url https://api.routergrowth.com/mcp, then codex mcp login routergrowth. Hermes: mcp_servers in config.yaml with auth: oauth (on a server, see the Hermes section). claude.ai and ChatGPT: add the URL as a connector. Each opens a RouterGrowth page where you sign in (password, Google, GitHub, or an emailed code) and approve a workspace. Details in the MCP quickstart.
• A key, for agents without a browser or without MCP (OpenClaw, CI, a script, an agent built on the Claude or OpenAI API): mint one on the dashboard under API keys, and the agent stores it with routergrowth keys add -k <key> -l main or sends it as the Bearer token over HTTP. Never commit the key.

The skill tells the agent to check for an existing MCP connection or stored credential before asking you for anything, and to verify access with the free balance call.

Docs your agent can read

Every page of this documentation has a markdown twin at the same URL plus .md (this page: /docs/quickstart-agents.md), and /docs/llms.txt indexes them all. The site-wide index at /llms.txt covers the catalog, providers and pricing.

So an agent needs no scraping and no browser: fetch llms.txt, follow the markdown, call the API.

Two tiers, one loop

discover returns capabilities and raw endpoints together, ranked, with hints that say when a capability wraps the endpoint an agent found. The skill teaches the agent to prefer the capability (normalized output, a quote, failover) and to go one tier down only when it needs a provider field the catalog does not expose. Raw endpoints has the full rules.

MCP

Prefer native tools over a skill file? Connect the MCP server at https://api.routergrowth.com/mcp: discover, inspect, run, batch_run, runs, get_run, history and balance as MCP tools. Claude Code, Codex, claude.ai and ChatGPT authenticate with OAuth in the browser; other clients send the same key as a header.

Spending rules the skill enforces

• Always inspect before the first run of a call: the schema and the price come from there, never from a guess.
• Respect routing.max_cost on every call, and set it on every raw endpoint run: with no quote, it is the only price control before the call.
• Stop and ask when a batch would exceed about $1.
• Report costs when they matter: when the user asked about price or set a budget, and as a total (the sum of each run's billing.charged) after a multi-run job.
• Never present sandbox mock data as real data.

Reporting feedback to the services you use

The services you call accept structured feedback through POST /feedback (or the submit_feedback tool). Your reports go to the engineers who build the service and are used to fix what got in your way. Use it.

Submit feedback when something prevented or complicated your task, even if you found a workaround: - a capability you needed does not exist (a filter, field, endpoint or bulk operation), so you made extra calls, filtered results yourself, or gave up; - the documentation was missing, ambiguous or wrong, and you had to guess; - a response was surprising or inconsistent with the docs; - an error message did not tell you how to fix the request; - the service behaved incorrectly; - latency, rate limits or pagination made the task impractical.

Before your final answer, check whether any of these happened. If one did and you have not reported it, report it now, then answer.

Don't submit feedback for your own mistakes that the service reported clearly, for things you have not tried, or more than once for the same problem in the same task.

Write it like this: - goal: what you or your user were trying to accomplish, in plain language. "Find companies currently hiring GTM engineers", not "call /companies/search". - message: what got in the way, specifically. "No hiring-role filter is available; q only matches company names." - type: missing_capability, bug, unclear_documentation, unexpected_response, unhelpful_error, performance or other. - Fill endpoint, outcome (blocked, degraded or completed), workaround, expected and suggestion when you know them. A concrete suggestion is the most useful thing you can give. - Reuse the same session_id for every report in one task.

Never include credentials, API keys, tokens, passwords, personal data, or your user's private content. Describe the shape of a request, not its secrets.

Keep going. Submitting feedback is best effort. Send it, then continue with the task or a workaround. If the response contains known_issue, the problem is already tracked: tell your user if it matters, and use the workaround if one is given.

Next

• Quickstart: CLI: the commands the agent will run.
• API overview: the HTTP surface behind them.

Questions this page answers

How can I add lead enrichment tools to my agent with a SKILL.md or MCP instead of writing wrappers?

Point the agent at https://www.routergrowth.com/SKILL.md. The skill teaches it the CLI and the HTTP interface, so contact.find, contact.verify, company.enrich, people.search and the rest of the catalog become things the agent can run, with the price quoted first and one balance behind them. If your harness speaks MCP, the MCP quickstart gives it the same catalog as native tools. Either way there is no wrapper to write per provider.

Which agents does the SKILL.md work with?

Claude Code, Codex, Cursor, Gemini CLI, Cline, GitHub Copilot, OpenClaw, Hermes and anything else that reads persistent instructions: the table above lists where the file lives for each. The skill is a plain markdown file following the Agent Skills convention, and the open-source skills repo adds workflow skills (cold email pipeline, hiring-signal outbound, AI visibility audit, brand mention sweep) on top of the core one.

Reading this as an agent? This page as markdown: /docs/quickstart-agents.md · every page: /docs/llms.txt