How to Connect Cohere to Real-Time SaaS Data with Unified.to MCP Server
August 28, 2025
Cohere's LLMs can power rich workflows — but most products still need access to live customer data from CRMs, ATSs, HRIS, or accounting systems. That usually means writing brittle glue code, normalizing APIs, and maintaining webhook jobs.
With Unified.to's MCP server, you can skip that complexity. Unified.to exposes 2300 SaaS integrations as real-time, callable tools that Cohere can use directly — no custom integration logic required.
In this guide, we'll walk through how to connect Cohere to Unified.to MCP so your application can give Cohere real-time access to customer SaaS data and actions.
Authentication
Tool calls to the Unified.to MCP server need a credential (listing tools does not). Send it in the Authorization: Bearer {token} header or the x-api-key header. Passing the token as a URL parameter is deprecated.
Unified.to documents five authentication options (as of 26 October 2026). For a server-side Cohere app, these two apply:
- Private (workspace key + connection)
Use your Unified.to workspace API key and add aconnectionparameter. This should never be exposed publicly. - Connection-scoped token
Generate a connection-scoped token with the Core MCP toolget_unified_connection_mcp_urlandmode=user; use the returnedrecommended.urland send the token in the Authorization header. The token grants access to one connection only, never exposes your workspace API key, and does not expire; to revoke it, delete the connection.mode=apiuses your workspace API key and must not be shared.
Integrating Unified.to MCP Server with Cohere
Cohere's chat API can directly consume tools defined by Unified.to MCP. Model names change often; use a current model from your provider's model list.
Manual Tool Orchestration:
- Call Unified's
/tools?type=cohereendpoint and pass the tool list to Cohere:import os import cohere import requests UNIFIED_API_KEY = os.environ["UNIFIED_API_KEY"] CONNECTION_ID = os.environ["CONNECTION_ID"] # Regional host without a path: US https://mcp-api.unified.to, EU https://mcp-api-eu.unified.to, AU https://mcp-api-au.unified.to UNIFIED_MCP_URL = os.environ.get("UNIFIED_MCP_URL", "https://mcp-api.unified.to") MODEL = "command-a-plus-05-2026" tools = requests.get( f"{UNIFIED_MCP_URL}/tools", headers={"Authorization": f"Bearer {UNIFIED_API_KEY}"}, params={"type": "cohere", "connection": CONNECTION_ID}, ).json() co = cohere.ClientV2(api_key=os.environ["COHERE_API_KEY"]) messages = [{"role": "user", "content": "Score this candidate for the Software Engineer job."}] response = co.chat(model=MODEL, messages=messages, tools=tools) while response.message.tool_calls: messages.append(response.message) for tc in response.message.tool_calls: result = requests.post( f"{UNIFIED_MCP_URL}/tools/{tc.function.name}/call", headers={ "Authorization": f"Bearer {UNIFIED_API_KEY}", "Content-Type": "application/json", }, params={"connection": CONNECTION_ID}, data=tc.function.arguments or "{}", ) messages.append({ "role": "tool", "tool_call_id": tc.id, "content": [{"type": "document", "document": {"data": result.text}}], }) response = co.chat(model=MODEL, messages=messages, tools=tools) print(response.message.content[0].text) - When Cohere requests a tool call, call Unified's
/tools/{id}/callendpoint. - Append the assistant message and a
toolmessage with thetool_call_idand the result, then callco.chatagain until Cohere stops requesting tools.
Example Prompt and Response
Prompt:
"Score this candidate for the Software Engineer job."
What happens:
- Your app passes Cohere the tools Unified MCP exposes for the connection.
- Cohere requests the tools that fetch the candidate and the job, and your app runs each call through
/tools/{id}/call. - Cohere scores the candidate from the returned data.
- Cohere returns a response like:
Response:
Candidate Jane Doe scored 92/100 for the Software Engineer job. Strengths: Python, distributed systems. Recommended for interview.
Advanced MCP Options
Unified.to MCP gives you granular control over how tools are exposed to Cohere:
permissions→ restrict available scopes.tools→ allowlist specific tool IDs.aliases→ adds extra descriptions to improve tool matching.hide_sensitive=true→ strips sensitive (PII) fields such as names, emails, phone numbers, addresses, date of birth, SSN/SIN and compensation, plus credential fields, from results.include_external_tools=true→ expose all vendor API endpoints, not just Unified.to's normalized models.
These options help you keep Cohere outputs predictable and secure in production-grade workflows.
Sample Snippet
import { CohereClientV2 } from 'cohere-ai';
// Unified connection ID that the tools should use
const connection = process.env.CONNECTION_ID;
// Regional host without a path: US https://mcp-api.unified.to, EU https://mcp-api-eu.unified.to, AU https://mcp-api-au.unified.to
const UNIFIED_MCP_URL = process.env.UNIFIED_MCP_URL || 'https://mcp-api.unified.to';
// Set to true to expose the vendor's own API endpoints as tools
const includeExternal = false;
const message = 'Score this candidate for the Software Engineer job.';
// Optional: list of specific tools that you want to use
const toolIds = ['get_messaging_message','list_messaging_message']
const cohereClient = new CohereClientV2({
token: process.env.COHERE_API_KEY || '',
});
const params = new URLSearchParams({
connection,
include_external_tools: includeExternal ? 'true' : 'false',
});
if (toolIds.length > 0) {
params.append('tools', toolIds.join(','));
}
// type selects the tool-list format and applies to GET /tools only
const toolsParams = new URLSearchParams(params);
toolsParams.set('type', 'cohere');
const tools = await fetch(`${UNIFIED_MCP_URL}/tools?${toolsParams.toString()}`, {
headers: { Authorization: `Bearer ${process.env.UNIFIED_API_KEY}` },
});
const toolsJson = await tools.json();
const completion = await cohereClient.chat({
model: 'command-a-plus-05-2026',
messages: [
{
role: 'user',
content: message,
},
],
tools: toolsJson,
});
for (const toolCall of completion?.message?.toolCalls || []) {
// call mcp server with toolCallId
const toolCallResponse = await fetch(`${UNIFIED_MCP_URL}/tools/${toolCall.function?.name}/call?${params.toString()}`, {
method: 'POST',
headers: {
Authorization: `Bearer ${process.env.UNIFIED_API_KEY}`,
'Content-Type': 'application/json',
},
body: toolCall.function?.arguments || '{}',
});
const toolCallResponseJson = await toolCallResponse.json();
console.log(JSON.stringify(toolCallResponseJson, null, 2));
}
Coverage and Infrastructure
Unified.to MCP runs as part of the Unified API:
- 242,210 real-time tools (growing weekly)
- 2300 integrations across 39 categories (ATS, CRM, HRIS, Accounting, Messaging, File Storage, and more)
- No stored customer records: Unified.to stores no customer records; logs hold operational metadata only
- Scoped security controls — permissions, aliases, and PII redaction
- Multi-region deployment — US, EU, and AU data centers for compliance
This ensures your Cohere workflows aren't just demos — they're secure, scalable, and designed for production-grade use cases.
Next Steps
Unified.to MCP works with major LLM providers, including OpenAI, Anthropic, Google Gemini and Cohere. That means you can build once and connect to any agent client.
Explore the MCP docs or book a demo to see it live in action.