# DiscoLike for agents One file for an agent setting up DiscoLike. Canonical URL: https://discolike.com/agents.md | Site summary: https://discolike.com/llms.txt | Human pages: https://discolike.com/agents/, https://discolike.com/mcp/, and https://discolike.com/api/ DiscoLike is a search engine over 82M+ crawled business websites. It returns companies ranked by what they actually do on the web, with firmographics, technographics, and contacts one call away. Requires a paid plan from $99/month; the person signs up and pays, the agent does everything else. ## Set up with one prompt > Set up DiscoLike using https://discolike.com/agents.md. Install the skill, connect this client, and help me sign in or create an account. Ask me before choosing a paid plan or running paid work. ## DiscoLike Prospecting Agent DiscoLike is building a managed prospecting agent that turns a company and persona brief into a run covering company discovery, ICP validation, contact qualification, and email finding. Describe the companies and people you want to reach. The agent coordinates the discovery and qualification steps, keeps partial results available, and shows why a run stopped. Use the app, CLI, API, or MCP. Availability depends on rollout to your environment. ### Costs and limits Managed prospecting uses existing processing charges and configured AI/search integrations. Agent decisions and contact qualification use customer AI credentials; brief interpretation is platform-funded. No bundled per-run price has been announced. Run limits bound the amount of work, not the dollar spend. Stopping a run preserves completed results and charges already incurred. CRM writeback and outreach are separate steps. ### Example brief Find UK sports retailers using Shopify, validate that they sell sports products, and find the Head of Ecommerce or founder at each accepted company with an email. Aim for 25 companies and one contact per company. ### Implemented interfaces (check availability before use) REST: POST /v1/prospecting/runs with Idempotency-Key creates a proposed plan. POST /v1/prospecting/runs/{run_id}/approve approves its plan_version. GET /v1/prospecting/runs/{run_id} returns progress and paginated results; DELETE stops new work. MCP: Get the person's approval for paid work before calling start-prospecting, which starts an auto-approved run. Reuse idempotency_key on retries. Poll get-prospecting-run every five seconds and paginate results; cancel-prospecting stops new work. Inspect stop_reason: completed does not necessarily mean the requested target was reached. Configure the required provider integrations before execution. Work on discovery, contacts, research, and email finding can incur charges; the company retrieval price is not the complete run cost. ## Pick an access mode | You are | Use | Why | |---|---|---| | An MCP client (Claude Code, Codex, Cursor, VS Code, Windsurf, Claude Desktop) | MCP server, section 2 | Discovery and research tools, OAuth in the browser | | Writing Python or shell | SDK and CLI, section 3 | Same OAuth browser login, JSON out, same credit balance | | Anything that speaks HTTP (n8n, Clay, Zapier, your own code) | REST API, section 4 | One GET returns ranked companies | | Any of the above | Install the skill first, section 1 | It teaches you when to use DiscoLike, how to search well, and how to keep spend down | ## 1. Skill (all agents) Source: https://github.com/Discolike/discolike-skills. The skill is the authoritative workflow: which mode to pick, how to count before you spend, how to run lookalike, phrase, and ICP searches, and how to open the account. Install the DiscoLike skill. If you're in Claude Code, run `claude plugin marketplace add Discolike/discolike-skills`, then `claude plugin install discolike@discolike`. If you're in another agent, run `npx skills add Discolike/discolike-skills --skill discolike` and select your agent. Use one installation method. You can read the skill directly at https://github.com/Discolike/discolike-skills/blob/main/skills/discolike/SKILL.md (raw: https://raw.githubusercontent.com/Discolike/discolike-skills/main/skills/discolike/SKILL.md). Then use the DiscoLike skill when working on this project. ## 2. MCP server - Server URL: `https://api.discolike.com/v1/mcp` - Transport: streamable HTTP. Auth: OAuth 2.1, browser authorization on first use, token stored by the client. Never ask the person for an API key for MCP. - Tools: discovery, company data, contacts, enrichment and analysis, workflow and crm. Full list: https://discolike.com/mcp.md ### Claude Code Terminal ```bash claude mcp add --transport http discolike https://api.discolike.com/v1/mcp ``` Run /mcp, pick DiscoLike, choose Authenticate. A browser window handles login. ### Codex Terminal ```bash codex mcp add discolike --url https://api.discolike.com/v1/mcp --oauth-resource https://api.discolike.com/v1/mcp ``` Codex opens the browser for OAuth on first use. ### Cursor .cursor/mcp.json ```json { "mcpServers": { "discolike": { "url": "https://api.discolike.com/v1/mcp" } } } ``` Settings, Tools, then Connect next to DiscoLike. ### VS Code .vscode/mcp.json ```json { "servers": { "discolike": { "url": "https://api.discolike.com/v1/mcp" } } } ``` VS Code prompts to authorize the first time Copilot calls a tool. ### Windsurf ~/.codeium/windsurf/mcp_config.json ```json { "mcpServers": { "discolike": { "serverUrl": "https://api.discolike.com/v1/mcp" } } } ``` Reload MCP servers in Windsurf settings after saving. ### Claude Desktop Settings, Connectors ```bash Add Custom Connector Name: DiscoLike URL: https://api.discolike.com/v1/mcp ``` Click Connect and authorize in the browser. ## 3. Python SDK and CLI ```bash pip install "discolike[cli]" discolike auth login # browser OAuth, or --api-key KEY discolike count --country US # structured count, not the ICP result count discolike discover --icp-prompt "B2B SaaS for logistics" --country US --max-records 100 --format json ``` Docs: SDK https://docs.discolike.com/sdk/, reference https://docs.discolike.com/sdk/reference/, CLI https://docs.discolike.com/cli/, source https://github.com/Discolike/discolike-python. `discolike auth login` opens the browser for OAuth and stores the token; the SDK picks it up, so no key is needed on a laptop. For headless pipelines, set `DISCOLIKE_API_KEY` instead; keys are issued after login at https://docs.discolike.com/api/access/#api-keys. ## 4. REST API - Base URL: `https://api.discolike.com/v1`. Header `X-API-Key` or an OAuth bearer token. - OpenAPI 3.1: https://api.discolike.com/v1/openapi.json. Examples and FAQ: https://discolike.com/api.md. Full reference: https://docs.discolike.com/ - `GET /count` is free for supported structured filters. It does not count semantic ICP/lookalike matches or predict the final qualified contact yield. Sample discovery results to check fit. ## 5. Account No account yet: one unauthenticated POST creates it, no credential comes back, the person confirms by email, logs in, and picks a plan. Then they authorize the MCP client or issue a key. ```http POST https://api.discolike.com/v1/public/signup Content-Type: application/json { "email": "jane@acme.com", "first_name": "Jane", "last_name": "Doe", "agent": "claude-code" } ``` Guide: https://docs.discolike.com/guides/agent-signup/. Auth summary: https://discolike.com/auth.md. ## 6. Rules that save the person money - Every search bills a query fee plus a fee per 1,000 new records. Records seen in the last 90 days are free. Rates: https://discolike.com/pricing.md - Count first. State the estimated cost before any pull larger than a sample, and get a yes. - Start at 100 to 500 records, check fit, then scale. Put every pull into an exclusion list so the next round returns only new companies. - Dead, parked, and redirected domains never appear in results and are never billed; every domain is re-validated about every 30 days. - Write raw results to disk as JSON or CSV. Chat output truncates. ## Prompts people paste Setup: Set up DiscoLike using https://discolike.com/agents.md. Install the skill, connect this client, and help me sign in or create an account. Ask me before choosing a paid plan or running paid work. Account: Create a DiscoLike account for me. Follow https://docs.discolike.com/guides/agent-signup/, ask me for anything you need, and tell me what to do next. Work: 1. Find 500 companies that look like stripe.com, adyen.com, and checkout.com. US and UK only, 50 to 500 employees. 2. List B2B companies whose homepage says "managed detection and response" and are headquartered in Germany. Count first, then pull 200. 3. Which of these 300 domains from our CRM use HubSpot and have grown headcount this year? Return a CSV. 4. Segment our 2,000 customers into ICP clusters, describe each, then find 1,000 lookalikes for the biggest cluster. 5. For the top 100 results, find the VP Sales or Head of Revenue with a verified email and push them to HubSpot.