> ## Documentation Index > Fetch the complete documentation index at: https://dev.jolts.xyz/llms.txt > Use this file to discover all available pages before exploring further. # Connect your AI assistant > Use the authenticated Streamable HTTP MCP interface. Configure a client that supports remote Streamable HTTP and custom bearer headers: Create a key using the [authentication guide](/authentication). An active or trialing subscription is needed for searches. MCP is the connection protocol that lets your agent discover and call Jolts' tools; it is not a separate data product. ```json theme={null} { "url": "YOUR_MCP_ENDPOINT", "headers": { "Authorization": "Bearer YOUR_API_KEY" } } ``` Copy the MCP endpoint from the marketing site's connection guide and replace `YOUR_MCP_ENDPOINT`. This is a connection description, not a universal client configuration file. Use your client's configuration format. OAuth discovery and authorization are not implemented; clients requiring OAuth-only connections are not yet supported. ## Set up your client Project configuration, private credentials, and connection checks. Copilot Chat setup with a secure API-key prompt. TOML configuration and private bearer-token setup. Project or global configuration for Cursor Agent. Remote HTTP configuration for the Cascade agent. Choose your assistant above for its exact configuration and verification steps. Keep your API key out of chat messages, screenshots, and source control. ### Other agents Look for **remote HTTP / Streamable HTTP**, an endpoint field, and an **Authorization header** field. Use `Bearer YOUR_API_KEY` as the header value. A client offering only OAuth login cannot use the current Jolts authentication. Do not paste the key into the endpoint URL or into the conversation. ## Ask your first question Once connected, try: > Use Jolts to find companies in Canada seeking warehouse equipment. Show the observed activity, > source and observation date for each match. Explain why it might matter, without claiming that > the company will buy. Tell me if coverage is limited or the search fails. The agent should call `search_companies`, not invent company research from this documentation. Searches consume your account's shared allowance. Review your client's approval prompt before running a search. A successful connection alone does not prove that your subscription allows it. ## Choose the right tool | Tool | Purpose | HTTP counterpart | | ------------------------ | -------------------------------------------------------- | --------------------------------------------- | | `search_companies` | Paginated company list; optional `background: true` | `POST /api/companies/search` | | `get_companies` | One to twenty known IDs or domains; current dataset only | `GET /api/companies?ids[]=…` or `domains[]=…` | | `get_usage` with `id` | Poll or replay an account-owned operation | `GET /api/usages/{id}` | | `get_usage` without `id` | Subscription limits and consumption | `GET /api/usage` | Discover exact input schemas using `tools/list`. For searches and company lookups, provide a unique `request_key` argument, or an `Idempotency-Key` header. The argument takes precedence. Tool discovery is unmetered; subscription checks happen inside the shared process. See the [tool reference](/mcp-tools) for arguments, filters, and response handling. ## Start and read a background company search ```json theme={null} { "jsonrpc": "2.0", "id": 1, "method": "tools/call", "params": { "name": "search_companies", "arguments": { "query": "Companies seeking EV charging equipment", "filters": { "countries": ["CA", "US"], "exclude_domains": ["example.com"] }, "limit": 10, "background": true, "request_key": "ev-company-list-1" } } } ``` Read `result.structuredContent.usage.id`, then call `get_usage` with that `id`. The `structuredContent` matches the HTTP business response. Treat `result.isError` and structured `error.code` as failures; an empty completed company list is a valid result with limited coverage, not a protocol error. ## Verify the connection After initialization, `tools/list` should advertise exactly `search_companies`, `get_companies`, and `get_usage`. Run the background example, then inspect the completed company's evidence. An API key can negotiate MCP without an active subscription but cannot run an entitled operation. Mintlify's documentation-search MCP, when hosted, searches these docs. It is not this authenticated company-data server. See the [HTTP tutorial](/quickstart) for credential and subscription troubleshooting. ## Test the transport directly If your client cannot connect, test in Bash with `API_BASE` and `API_KEY` from the [quickstart](/quickstart). This negotiates the protocol without running a metered search: ```bash theme={null} curl --fail-with-body "$API_BASE/mcp" \ -H "Authorization: Bearer $API_KEY" \ -H "Content-Type: application/json" \ -H "Accept: application/json, text/event-stream" \ -d '{"jsonrpc":"2.0","id":1,"method":"initialize","params":{"protocolVersion":"2025-11-25","capabilities":{},"clientInfo":{"name":"docs-check","version":"1.0"}}}' ``` Expect `result.serverInfo` and a negotiated protocol version. Agent clients normally handle this handshake themselves. The server is stateless; it does not provide a server-pushed event subscription. ## Troubleshooting * **Cannot connect:** use the application host's `/mcp`, not the documentation-search MCP URL. * **Unauthorized:** check that the client sends the bearer header on every request. * **Tools appear but searches fail:** inspect `result.isError` and the structured error code; a valid key can discover tools without an active subscription. * **Missing request key:** add a distinct `request_key` to each search. Reuse it only for identical retries. * **OAuth prompt only:** that client configuration is not supported yet. Use a remote HTTP client that accepts custom bearer headers.