# Deep Search directory connectors

Research companies and vehicles using your existing Deep Search account and included usage.

## Connect

Use Streamable HTTP with OAuth 2.0 authorization code + PKCE (S256).
- ChatGPT: https://deepsearch.app/api/mcp/chatgpt
- Claude: https://deepsearch.app/api/mcp/claude
- Muse directory mode: https://deepsearch.app/api/mcp/muse (while directory mode is enabled; the existing custom connector otherwise has a broader catalogue)
- Authorization-server metadata: https://deepsearch.app/.well-known/oauth-authorization-server
- Resource metadata: append the endpoint path to https://deepsearch.app/.well-known/oauth-protected-resource
- Permissions: search (company search), dossier (company and vehicle reports).

Each connection is bound to its own endpoint. A token for one connector cannot access another connector or the general REST/MCP API. Sign in to your own Deep Search account at the consent screen. Disconnect through Deep Search Developers > Integrations > Connected apps to revoke the connection. Availability in a directory depends on that platform's review.

## Tools

### Find companies

search_companies: Find distinct companies by name or business domain, with public sources. This starts research and may consume existing included usage. Keep entities and jurisdictions separate. No people or contact lookups. Retry with the same requestKey.

Input JSON Schema:

```json
{
  "$schema": "http://json-schema.org/draft-07/schema#",
  "type": "object",
  "properties": {
    "query": {
      "type": "string",
      "minLength": 1,
      "maxLength": 240,
      "description": "Company name or business domain to research."
    },
    "requestKey": {
      "type": "string",
      "pattern": "^[\\w.:-]{1,160}$",
      "description": "A unique request key. Retain it on retries; use a new key for a different lookup."
    }
  },
  "required": [
    "query",
    "requestKey"
  ],
  "additionalProperties": false
}
```

### Research a company

build_company_dossier: Build a sourced report on one company. Supply its name and, when known, its business domain and jurisdiction to distinguish entities. Uses existing included usage; a cached report is free. Personal contact enrichment and sensitive research are excluded. Retry with the same requestKey.

Input JSON Schema:

```json
{
  "$schema": "http://json-schema.org/draft-07/schema#",
  "type": "object",
  "properties": {
    "name": {
      "type": "string",
      "minLength": 1,
      "maxLength": 240,
      "description": "The selected company's name."
    },
    "domain": {
      "type": "string",
      "maxLength": 255,
      "pattern": "^(?:[a-z0-9](?:[a-z0-9-]*[a-z0-9])?\\.)+[a-z]{2,}$"
    },
    "jurisdiction": {
      "type": "string",
      "minLength": 1,
      "maxLength": 120
    },
    "requestKey": {
      "type": "string",
      "pattern": "^[\\w.:-]{1,160}$",
      "description": "A unique request key. Retain it on retries; use a new key for a different lookup."
    }
  },
  "required": [
    "name",
    "requestKey"
  ],
  "additionalProperties": false
}
```

### Research a vehicle VIN

lookup_vehicle: Decode a valid VIN and return available specifications, recalls and vehicle-history evidence with sources. Uses existing included usage; a cached report is free. Does not identify owners or track a vehicle. Missing records do not establish a clean history. Retry with the same requestKey.

Input JSON Schema:

```json
{
  "$schema": "http://json-schema.org/draft-07/schema#",
  "type": "object",
  "properties": {
    "vin": {
      "type": "string",
      "minLength": 17,
      "maxLength": 32
    },
    "requestKey": {
      "type": "string",
      "pattern": "^[\\w.:-]{1,160}$",
      "description": "A unique request key. Retain it on retries; use a new key for a different lookup."
    }
  },
  "required": [
    "vin",
    "requestKey"
  ],
  "additionalProperties": false
}
```

## Examples

- Find OpenAI as a company; distinguish entities and jurisdictions before building a company report.
- Research Microsoft Corporation using microsoft.com as the business domain and cite the company evidence.
- Decode VIN 1HGCM82633A004352 and explain available recall/history evidence and its gaps.

## Scope and data

These connectors expose only the three tools above. They do not expose people, phone, email, username, reverse-image or photo-location lookups, personal reports, generic job executors, account settings or purchases. Company reports may include public professional information about company executives; they do not build personal dossiers. The research engine excludes sensitive enrichment, face matching, breach/account probes, personal-life data, precise personal addresses and Meta-platform sources. Its restricted company cache is separate from the general product cache.

Lookups may consume existing included usage. No subscription, credit purchase or automatic top-up can be made here. If usage is unavailable, the tool reports this without initiating a purchase. Cached reports do not consume new usage.

## Results, retries and failures

All tools start stateful research and declare readOnlyHint=false, destructiveHint=false, openWorldHint=true. Supply a unique requestKey and retain it on retries. A request key is scoped to the connector, account and operation. A repeated key with different arguments is rejected. An uncertain or running request does not start a second provider call. Completed results are returned as text and structured JSON; no UI, uploaded files, skills or saved personal-report resources are exposed.

An empty candidates array is a completed company search with no matches. A provider failure is an error, not an empty result or a clean vehicle report. Invalid arguments, expired/revoked credentials, missing permissions, rate limits and unavailable included usage have distinct error codes. Do not change the requestKey to retry an uncertain outcome. Contact support if a request remains processing.

Limits: 120 requests/minute per credential, 1,200 per egress IP, and at most 10 research calls/minute per credential (a lower credential limit is respected). Company research may take several minutes; host timeout and recovery behavior must be tested in the intended client.

Privacy: https://deepsearch.app/privacy
Terms: https://deepsearch.app/terms
Support/security: support@deepsearch.app; https://deepsearch.app/contact