AI agents are like autonomous coworkers — software entities capable of reasoning, planning, and acting on your behalf by calling external tools.
Agentic AI is the architectural approach behind this shift: systems where agents orchestrate and chain together actions to solve problems without constant human guidance.
This is where the Model Context Protocol (MCP) fits in. MCP is a new and emerging protocol that provides a standardized way for these agents to discover and call external tools — like Coveo’s Answer API, Search API or Passage Retrieval API.
Introducing the Coveo MCP Server: A Community Accelerator for Agentic Innovation
The Coveo MCP server is a community sample that we’re offering as-is for exploratory use. It’s being shared early as an accelerator to help our customers and partners experiment with Agentic frameworks.

With the Coveo MCP Server, you can now plug those same capabilities into many leading LLM clients, frameworks, and platforms — Salesforce Agentforce, AWS Bedrock Agents, OpenAI Agents SDK, LangChain, Claude, or your own custom-built logic with the Model Context Protocol (MCP) without needing custom connectors, wrappers, or prompt hacks.
Together, agentic AI and the MCP Server represent the future of AI integration: modular, scalable, and standards-based. They let you build smarter systems faster — ones that can reason, retrieve, and respond using the best tools available, including Coveo.
Before an AI agent can reason, respond, or take action, it needs access to relevant, accurate, and secure information. This is the fundamental challenge that Coveo solves, making it an essential component—not an optional one—in any serious agentic AI architecture. For organizations building AI agents that operate on complex, high-stakes content, Coveo provides the critical retrieval layer that grounds the agent in your business reality. It ensures that the information fueling your agent is not just accessible, but also relevant, permission-aware, and tailored to the user’s context, dramatically reducing hallucinations and enabling reliable, scalable, and secure outcomes.
Why are we doing this? The landscape is evolving rapidly. Standards like MCP are gaining traction fast, and we believe that following the industry’s momentum is the best way to support our customers. We’re also tracking adjacent frameworks — OpenAI function-calling, LangChain’s tool schema, AWS Bedrock’s agent tool interface, and Google’s A2A protocol for agent-to-agent communication — so we can add support where our customers need it. As an agnostic retrieval provider, we’re building for compatibility and scalability. Our approach is simple: meet customers where they are and future-proof their integrations by aligning with open, widely accepted protocols.
Coveo APIs: Your AI Agent Building Blocks
If you’re already using Answer API or the Passage Retrieval API, your agents already have a rich foundation to build on. Here’s a quick refresher on what these APIs offer:
API | What it does | How agents use it |
Passage Retrieval API | Returns only the most relevant text snippets (“passages”) from large documents | Helps agents find pinpoint answers without prompting over full documents |
Answer API | Produces generative answers using retrieved content, with citations | Enables agents to respond in a complete, human-readable way |
Search API | Performs full-text search across all your indexed content (e.g., SharePoint, Salesforce, Zendesk) | Useful for surfacing documents, FAQs, or records that match a query |
Each of these APIs is production-tested, privacy-aware, and secure by design—key requirements when building enterprise-ready agents.
What does the Coveo MCP Server do?
Out of the box, it gives you:
- MCP Tools for Search, Retrieval and Answering
Agents can search indexed enterprise content and extract the most relevant passages—directly, without prompt engineering. Agents can also get generated natural-language answers with citations or resolve full support requests using Coveo’s Relevance Generative Answering via the Answer API. - A Launchpad for Custom Tooling
Want to add personalization, recommendations, or even invoke backend workflows? You can easily extend the boilerplate to expose your own MCP tools.
Use Cases for Coveo Customers
Here’s how the Coveo MCP Server can help you expand what you’re already doing:
You already use… | With the Coveo MCP Server, you can now… |
CRGA or Answer API in your support portal | Build an autonomous support agent that answers and deflects tickets |
Passage Retrieval for knowledge access | Create an internal agent that summarizes docs and policies for employees |
Basic Coveo search integration | Build a research assistant that explores, filters, and reasons over results |
A Chatbot UI | Replace static answers with dynamic, multi-step workflows backed by your content |
Try it yourself
Getting started is simple:
- Clone the repo: github.com/coveo-labs/coveo-mcp-server
- Install and setup: Full instructions in the README – https://github.com/coveo-labs/coveo-mcp-server/blob/main/README.md
- Configure the environment:
bash cp .env.example .env # Set COVEO_API_KEY, COVEO_ORGANIZATION_ID, COVEO_ANSWER_CONFIG_ID
- Run locally or deploy in a secure environment
Testing with MCP Inspector
You can test the server with the MCP Inspector in two ways:
- Using the mcp dev command:
bash mcp dev /[path to project]/src/coveo_mcp_server/__main__.py
This will start the MCP Inspector at http://localhost:5173, where you can interact with and test the server’s functionality.
- Using npx with the MCP Inspector:
bash npx @modelcontextprotocol/inspector \ /[path to]/uv \ --directory /[path to project]/coveo-mcp-server \ run \ --with mcp \ mcp run src/coveo_mcp_server/__main__.py
- Connect your favorite agent framework: Amazon Bedrock Agents, LangChain, Claude Desktop, GPT agent, or others
{ "mcpServers": { "coveo": { "command":"/[path to]/uv", "args" : [ "--directory", "/[paty to project]/coveo-mcp-server", "run", "src/coveo_mcp_server/__main__.py" ] } } }
No prior experience with MCP or agent frameworks is needed to test basic workflows.
Prerequisites:
A Coveo org with your enterprise content as the data from which to generate answers. The content that you choose to use for answer generation and the quality of that content have a direct impact on the quality of the answers that are generated.
See Relevance Generative Answering content requirements and best practices, and Passage Retrieval content requirements and best practices for more information.
Search API
The Coveo Search API is the foundational service that enables querying your organization’s unified index. It processes search requests, applies relevance tuning, and returns results tailored to the user’s context and permissions. The API supports complex search queries with filters, facets, and sorting options, and integrates with Coveo’s machine learning models to enhance result relevance based on user behavior and content. Security is a key feature, ensuring that users only access content they are authorized to view, respecting your organization’s security protocols. Additionally, the Search API allows for query pipeline configurations to tailor the search experience to specific use cases. To perform a search query, send a POST request to the Search API endpoint with the desired parameters.
Coveo Passage Retrieval API
Coveo’s Passage Retrieval model retrieves the most relevant segments of text (aka passages) from your dataset for a natural language user query that goes through a Coveo query pipeline. Passage Retrieval selects and feeds relevant passages to your enterprise’s Retrieval-Augmented Generation (RAG) system to enhance your large language model -powered applications. Therefore, in the context of your RAG system implementation, the Passage Retrieval model retrieves the most relevant passages, and the Passage Retrieval API provides the passages to your enterprise RAG system.
Answer API
Answer API is an extension of our managed solution Coveo Relevance Generative Answering (CRGA) for custom UIs or applications. The CRGA model generates answers to complex natural language user queries in a Coveo-powered search interface. This model leverages generative AI technology to generate an answer based solely on the content that you specify, such as your enterprise content. The content resides in a secure index in your Coveo organization. Relevance Generative Answering works with Coveo’s existing indexing, AI, personalization, recommendation, machine learning, relevance, and security features. Resulting in a powerful enterprise-ready solution that generates answers that are relevant, personalized, and secure. All while respecting your enterprise’s privacy and security.
API Key or Search Token
A valid Coveo API key for anonymous search – To allow unauthenticated users to perform search queries in your implementation, you must configure and use an API key with the necessary privileges. This key must be created in the Coveo Administration Console using the “Anonymous search” template, or be manually configured to include the appropriate privileges for query execution and (optionally) analytics data push. For best practices and implementation guidance, see the following documentation:
https://docs.coveo.com/en/105/build-a-search-ui/use-api-key-authentication-with-the-search-api and https://docs.coveo.com/en/1718/manage-an-organization/manage-api-keys
Demos — Examples
Claude Desktop (claude.ai/download):



Cursor (cursor.com):



Windsurf (windsurf.com)


Visual Studio Code (code.visualstudio.com):


What’s next?
The Coveo MCP Server is provided as-is, intended purely for educational and exploratory purposes. This is not a production-ready product, but an early-stage project designed to help you start experimenting with generative AI in a hands-on way.
As the landscape for AI agents and agentic applications rapidly evolves, we are committed to our long-standing position as an agnostic, industry-leading platform. Our goal is to support the diverse range of open standards and communication protocols used by the main agentic platforms. This ensures you can connect your AI agents to real, enterprise-grade data, regardless of the frameworks you choose.
We encourage you to use this lightweight starter to:
- Integrate Coveo APIs into LLM-based agent workflows
- Prototype intelligent copilots or automated support flows
- Explore the full potential of RAG in your enterprise context
Interested in going further?
Coveo is actively monitoring industry trends and adoptions as we work towards delivering robust, enterprise-grade solutions for the agentic future. We’re always eager to collaborate and help you unlock the next stage of your AI journey. Reach out to your Coveo Customer Success Manager, Account Executive, or Sales Representative to explore a partnership with us.
Online Resources
- GitHub – Coveo MCP Server
- Model Context Protocol
- Model Context Protocol Specification (2025-03-26)
- For Server Developers – Model Context Protocol
- The official Python SDK for Model Context Protocol servers and clients
- Coveo Passage Retrieval API
- Coveo RGA / Answer API
- Coveo Search API
- About Semantic Encoder (SE) | Coveo Machine Learning
- About Relevance Generative Answering (RGA) | Coveo Machine Learning
- About Coveo Passage Retrieval (CPR)
- Unlock the Power of Precise Information: Introducing the Coveo Passage Retrieval API
- Manage API keys | Coveo Platform
- Search API authentication methods | Coveo Platform