Lev8 MCP

Lev8 MCP lets your AI assistant search entities and contacts, answer research questions, enrich records, extract public lists, and manage credits through one hosted endpoint. There is no local server, executable, or package to install.

https://app.lev8.com/mcp

Create an API Token and set it as an environment variable, then click to install:

export LEV8_API_KEY="lev8_live_..."

Add to Cursor

Installs the config. Cursor reads LEV8_API_KEY from your environment at runtime.

Add to VS Code

Installs the config. VS Code prompts for the Token on first connection.

Connect your client

If you haven’t created a Token yet, follow the Authentication guide. Store the complete lev8_live_... value as LEV8_API_KEY in your environment.

Never commit your Token to source control. The values below are placeholders.

Codex
Claude Code
Gemini CLI
Other

export LEV8_API_KEY="lev8_live_..."

codex mcp add lev8 \
  --url https://app.lev8.com/mcp \
  --bearer-token-env-var LEV8_API_KEY

Confirm the server is registered with codex mcp list. Restart Codex after adding it so the new Tools are loaded. The ChatGPT desktop app, Codex CLI, and Codex IDE extension share this configuration on the same host.

export LEV8_API_KEY="lev8_live_..."

claude mcp add-json --scope user lev8 \
  '{"type":"http","url":"https://app.lev8.com/mcp","headers":{"Authorization":"Bearer ${LEV8_API_KEY}"}}'

The single quotes preserve ${LEV8_API_KEY} for Claude Code to resolve when it connects. Run claude mcp get lev8 to inspect the saved configuration, then use /mcp inside Claude Code to check the connection.

export LEV8_API_KEY="lev8_live_..."

gemini mcp add --scope user --transport http \
  --header "Authorization: Bearer ${LEV8_API_KEY}" \
  lev8 https://app.lev8.com/mcp

Run gemini mcp list to check the connection. Gemini CLI expands the environment variable when the command runs and saves the resulting header in your private user settings, so do not share that file.

Lev8 MCP works with any client that supports remote Streamable HTTP servers. Point the client at https://app.lev8.com/mcp and send the Token in the Authorization header:

Authorization: Bearer lev8_live_...

The config key for the URL varies by client:

Client	Config location	URL key
Cursor	`~/.cursor/mcp.json`	`url`
VS Code	`.vscode/mcp.json` (Command Palette > MCP: Open User Configuration)	`url`
Windsurf	`~/.codeium/windsurf/mcp_config.json`	`serverUrl`
Zed	Zed `settings.json` (under `context_servers`)	`url`
Warp	Settings > MCP Servers > Add MCP Server	`url`

Most clients use the standard mcpServers shape:

{
  "mcpServers": {
    "lev8": {
      "type": "http",
      "url": "https://app.lev8.com/mcp",
      "headers": {
        "Authorization": "Bearer YOUR_LEV8_API_KEY"
      }
    }
  }
}

The Token is request-scoped. It is not a Tool input and does not appear in Tool schemas or results.

Available Tools

entity_search_create: Create a billed asynchronous search for companies, people, creators, Shopify stores, Xiaohongshu, or WeChat Official Accounts.
entity_search_status: Check the progress of an entity search task.
entity_search_fetch: Read one page of results from a completed or in-progress search.
contact_search: Find an email or phone number. Only a non-empty result is charged.
deep_search: Answer an open-ended research question synchronously.
enrich: Generate structured fields from entity context and a caller-provided JSON Schema.
extract: Extract social, GitHub, or public web-page lists synchronously.
credit_balance: Read the available credits for your billing account.

See the Entity Search, Contact Search, Deep Search, Enrich, Extract, and Credit balance API references for full parameter details.

Try it

After the Tools appear in your client, ask it to use Lev8 in natural language:

Example prompts

Search for entities

Use Lev8 to find 20 AI infrastructure companies in the United States.

Calls entity_search_create, then entity_search_status and entity_search_fetch automatically.Find a contact

Find the work email for the VP of Sales at Example Company using Lev8.

Calls contact_search and returns the result directly.Research a question

Use Lev8 Deep Search to explain who founded Mercury and distinguish the companies that share that name.

Calls deep_search. Its answer is a dynamic, non-null JSON value and may include a generated result plus citations.Enrich an entity

Use Lev8 to enrich this company record with a summary and current CEO: Company: OpenAI, Website: https://openai.com.

Calls enrich with an object JSON Schema for the requested fields.Extract a public list

Use Lev8 to extract company names and websites from https://example.com/directory.

Calls extract with type: web2list, the public URL, and an extraction instruction.Check credits

Check my Lev8 credit balance.

Calls credit_balance.

Your client chooses the relevant Tool and asks you to approve the call according to its own permission settings.

Entity Search workflow

Entity Search is asynchronous. Create a task, monitor it, then fetch results in pages.

Create

Call entity_search_create with an entity_type and objective. For xiaohongshu searches, or WeChat Official Account searches with entity_type: "wechat", set enable_web_search to true. Save the returned leads_search_id.

Monitor

Call entity_search_status until the state becomes done or error. You can begin fetching as soon as ready_count is positive. Xiaohongshu and WeChat Official Account tasks usually take 1 to 2 minutes before results are ready.

Fetch

Call entity_search_fetch. Advance pts by the returned count until all ready_count results have been read.

A task with status: error can still have ready results. Fetch them before closing the task.

Contact Search workflow

Contact Search is synchronous and returns the result in a single call.

Call contact_search with a contact_type (email or phone) and an objective describing the person and company. The result is returned directly.

Only a non-empty result is charged. If no contact is found, no credits are deducted.

Synchronous tools

deep_search, enrich, and extract each return a result in one Tool call. Deep Search passes through its answer as dynamic JSON; do not assume it is always a string or that an object has fixed fields. Enrich returns an object shaped by the supplied JSON Schema. Extract returns an array on success and a failure-reason string when success is false. For web2list, extract requires an instruction and a public HTTP or HTTPS URL. Social and GitHub extraction limits default to 100 and can be at most 10000; web2list normalizes the limit to 100.

Billing and retries

Every Tool call goes through the Lev8 API, so Token scopes, credit billing, and idempotency handling still apply. The MCP server generates an internal idempotency key for every paid POST Tool: entity_search_create, contact_search, deep_search, enrich, and extract. That key is not exposed as a Tool input.

The first release does not automatically retry paid operations. Calling the Tool again creates a new key and is treated as a new billable operation. Do not repeat a call whose outcome is unknown.

Troubleshooting

The server returns 401 Unauthorized

Confirm the Token is complete, active, and sent as Authorization: Bearer <token>.

Authorization: Bearer lev8_live_...

Tools do not appear

Restart your client after changing its MCP configuration. Some clients require a full restart to detect new MCP servers.Check the MCP server list or connection logs in your client to confirm lev8 is registered.

A Tool reports a scope error

Enable the required scope on your Token from API Tokens, or create a new Token with the correct access.

A paid Tool times out

Do not immediately call it again. The original operation may still have consumed credits. Check with entity_search_status or credit_balance before retrying.

Cursor cannot read the Token

If Cursor was opened from the macOS Dock or another desktop launcher, that process may not have access to LEV8_API_KEY.Launch Cursor from a terminal where the variable is exported, or configure the variable in your desktop environment.

Transport details

Lev8 MCP uses Streamable HTTP (POST with JSON). The endpoint is stateless and does not create persistent sessions or send server-initiated notifications.
Authentication uses the lev8_live_... API Token as a Bearer credential. The HTTP API and MCP use the same Token.
OAuth, dynamic client registration, and browser authorization are not supported in the first release.
The MCP endpoint does not provide Token creation, listing, or revocation Tools.
Objective text, results, contact details, and Tokens are not written to MCP logs.

Next steps

Agent Skills

Install the Lev8 skill for Codex, Claude Code, or Cursor.

Authentication

Create, store, and rotate API keys.

Entity Search

Review entity types, enrichment fields, and deduplication options.

Contact Search

Find email addresses or phone numbers.

Deep Search

Answer open-ended research questions.

Enrich

Generate structured fields for an entity.

Extract

Extract social, GitHub, and public web-page lists.

Getting started

API reference

MCP

API behavior

Add to Cursor

Add to VS Code

Connect your client

Available Tools

Try it

Entity Search workflow

Contact Search workflow

Synchronous tools

Billing and retries

Troubleshooting

Next steps

Agent Skills

Authentication

Entity Search

Contact Search

Deep Search

Enrich

Extract

Add to Cursor

Add to VS Code

​Connect your client

​Available Tools

​Try it

​Entity Search workflow

​Contact Search workflow

​Synchronous tools

​Billing and retries

​Troubleshooting

​Next steps

Agent Skills

Authentication

Entity Search

Contact Search

Deep Search

Enrich

Extract

Connect your client

Available Tools

Try it

Entity Search workflow

Contact Search workflow

Synchronous tools

Billing and retries

Troubleshooting

Next steps