Conga Launches an MCP Server That Lets Claude and Copilot Act on Quotes and Contracts: What It Means for AI Agents in B2B Order Flows
Conga, the quoting and contracting software company, launched an MCP server that lets Claude and Copilot check approval status and send agreements for signature. Which B2B deal steps are now open to AI agents, what has not been disclosed, and what to check in your own order flow.
Key Takeaways
- On October 14, 2026, Conga, which sells quoting, pricing, and contracting software, launched an MCP server. Claude and Microsoft Copilot can now work with Conga data directly, from checking where a quote stands in approval to sending the agreement for signature, without opening Conga
- Most commerce MCP servers so far have opened the shopping steps themselves: product search, carts, and orders. Conga opens the steps that come before the order in a B2B deal: pricing, quoting, approvals, and contracts
- B2B e-commerce teams, manufacturers, and distributors should now sort their order flow into steps AI may read and steps AI may execute. Pricing and permission details have not been disclosed, so any adoption decision needs direct confirmation
Working on quotes and contracts without opening Conga
Conga announced a wave of AI-first product innovations at Conga Connection in Amsterdam, headlined by the launch of the Conga Model Context Protocol (MCP) Server.
finance.yahoo.com"Which contracts expire in the next 90 days?" A CFO asks Copilot, and the answer comes back without anyone opening the contract management system. That is the scene at the center of the press release Conga issued on October 14, 2026, at Conga Connection, its annual event in Amsterdam.
The headline is the Conga MCP Server. MCP (Model Context Protocol) is an open standard for connecting AI assistants to business software. Conga built an entry point that follows this standard, so Claude, Microsoft Copilot, and other MCP-compatible assistants can query data on the Conga Advantage Platform in real time.
The release gives three examples. A CFO asks which contracts are about to expire. A legal operations lead asks Claude what an amendment changed. And a sales representative checks where a quote stands in approval and then sends the agreement for signature. Only the third example goes beyond reading information; it moves the business process forward.
Where Conga sits in the deal
Conga's products cover the stretch of a B2B deal from the decision to sell until the contract is signed. The lineup includes CPQ (Configure, Price, Quote), which combines specifications and prices into quotes; CLM (Contract Lifecycle Management), which manages agreements from drafting through signature and renewal; price optimization; rebate management; and document generation for quotes and contracts.
According to the release, Conga has more than 10,000 customers worldwide, including over half of the Fortune 100. Its contract management also comes in versions that run on Salesforce, so sales teams can manage opportunities in the CRM and build quotes and contracts from there.
Conga calls these connected steps the "Commerce Chain." The idea is to reconnect pricing, quoting, contracting, rebates, and documents that otherwise run in separate systems. The new MCP server is the doorway that opens that connected chain to outside AI assistants.
What the MCP server opens, and what it does not yet
MCP itself is simple. The MCP specification defines three parties: the AI application (the host), the client inside it that handles connections, and the server that provides data and capabilities. A server can offer data for the AI to read (resources), functions for the AI to execute (tools), and templated instructions (prompts). A software vendor that builds one server can use the same entry point for Claude and for Copilot.
What stands out in Conga's case is that it opens both reading and acting. The three examples in the release break down as follows.
| User | Assistant | Request | Type of operation |
|---|---|---|---|
| CFO | Microsoft Copilot | List contracts that expire in the next 90 days | Read |
| Legal operations lead | Claude | Explain what an amendment changed | Read |
| Sales representative | Not specified | Check where a quote stands in approval and send the agreement for signature | Read and act |
Conga's blog post on the October release lists contracts, quotes, obligations, and approvals as the information available through the MCP server. Obligations here means what a contract commits to, such as delivery dates, payment terms, and renewal deadlines. These are exactly the items that sales, legal, and finance check again and again before an order closes.
Sending an agreement for signature moves a deal toward completion. Send the wrong version and it lands in front of the counterparty ready to sign. The MCP specification says hosts must obtain explicit user consent before invoking any tool. The specification also admits that MCP cannot enforce these principles at the protocol level, so the real safeguards depend on the assistant's confirmation screens and on how Conga designs permissions.
It also helps to separate the MCP server from AiMe, Conga's built-in AI. AiMe is a conversational AI used inside Conga's own screens, and it has now grown to more than 90 tools and 19 skills. The opening of the release puts it at nearly 100 tools. Skills include redline suggestions, search within contracts, and screening for GDPR, HIPAA, and sanctions exposure. In Conga's framing, AiMe is the intelligence inside Conga, and the MCP server is the doorway that takes Conga data out to other assistants. They play different roles.
What the MCP server exposes to outside assistants, however, has not been published as a list. The public materials do not say whether all 90+ AiMe tools are available from outside, or whether access is limited to reading plus a few actions such as sending for signature.
How it compares with other commerce MCP servers
Conga is not the first to open business systems through MCP. Microsoft made MCP support in Copilot Studio generally available in May 2025, and Salesforce laid out its MCP support for developers in June 2025. The commerce platform commercetools offers Commerce MCP, which handles carts and orders, in both managed and self-hosted versions.
| Provider | Role in MCP | Main process covered |
|---|---|---|
| Conga | MCP server (launched October 2026) | Pricing, quotes, approvals, contracts, contractual obligations |
| commercetools | Commerce MCP (managed and self-hosted) | Carts, orders, and other commerce resources |
| Salesforce | MCP servers, plus connections from Agentforce | CRM data and workflows (Conga's contract management also supports Salesforce MCP) |
| Microsoft | MCP support in Copilot Studio (generally available May 2025) | The connecting side (client) for outside MCP servers |
Side by side, the difference is clear. Most commerce MCP servers have opened the steps of finding products, adding them to a cart, and creating an order. For consumer shopping, that is the heart of the transaction.
B2B deals work differently. An order comes only after quotes with hundreds of lines, discount approvals, rebate calculations, and contract negotiation. In this release Conga also added CPQ capabilities that check every line of quotes with hundreds of lines against the same pricing policy and margin rules, and calculate rebates line by line in real time. What the MCP server opens is this stage of reaching agreement before the order.
The Salesforce integration points in the same direction. Conga CLM for Salesforce and Conga Contracts for Salesforce now support Salesforce MCP, letting users create, generate, review, negotiate, sign, publish, and manage contracts through natural language requests. On the procurement side, Conga announced an SAP Ariba integration that automatically creates a contract in Conga CLM when a sourcing event is awarded in Ariba, and publishes the signed agreement back to Ariba. On both the selling and buying sides, the design lets contracts flow between systems without manual handoffs.
What has not been disclosed, and why caution is warranted
Several pieces of information needed for an adoption decision are missing from the announcement.
- Pricing for the MCP server, and whether it is included in existing licenses: not disclosed
- Availability (general availability or limited to some customers, and which Conga products are covered): not disclosed
- The list of tools exposed to outside assistants, and which actions besides sending for signature can be executed: not disclosed
- How each user's permissions map to view and edit permissions inside Conga: not disclosed
Permissions deserve particular attention. The MCP security best practices warn against granting broad permissions up front and recommend starting with minimal scopes and expanding step by step. Contract and pricing data sit at the core of a company's confidential information. Rolling this out company-wide without clarity on whose assistant can act on which contracts, and how far, is risky.
The broader market also calls for a cool head. In June 2025, Gartner predicted that over 40% of agentic AI projects will be canceled by the end of 2027, citing escalating costs, unclear business value, and inadequate risk controls. Gartner also pointed to "agent washing," the rebranding of existing chatbots and similar products as agents, and estimated that only about 130 of the thousands of agentic AI vendors are real.
Two of Conga's three examples simply find information and answer questions. Saving search time has value, but it is still some distance from agents that move deals forward on their own. How far actions such as sending for signature are delegated, and under which approval rules, will be the real test of what this can do.
What B2B e-commerce teams, manufacturers, and distributors should check
Even companies that do not use Conga can treat this announcement as a prompt to review their own order flow. Write out every step from quote request to order, then sort each step into one of three groups. The order in which to bring in AI agents becomes visible.
The first group is steps where reading alone delivers value. Checking contract expiry dates, quote approval status, and customer-specific price terms does not move a deal even if the AI gets it wrong, and the benefit is easy to see. The second group is steps where AI drafts and a person finalizes. Building quotes and proposing contract edits belong here. The third group is steps where AI executes, such as sending for signature or confirming an order. For these, rules such as approval limits and eligible customers must be fixed in the system first.
Another trend is easy to miss: agents are also entering at the front door of the quote. The intake agent that commercetools built with Mirion Technologies, a maker of nuclear measurement and detection equipment, turns order requests arriving by email, PDF, or spreadsheet into quotes and carts, and hands them to sales reps as drafts to review. Agents from different products are starting to cover each step, from receiving a request to quoting, approval, and contracting. That is what B2B looks like today.
The question then becomes whether price lists and contract terms are in a form machines can read. If price lists live in Excel files and contracts sit as PDFs scattered across individual reps, building an MCP entry point leaves nothing to pass through it. Organizing where data lives and who can access it comes before choosing any tool.
Conclusion
Conga's MCP server opens the quote, approval, and contract steps that come before the order in a B2B deal to AI assistants. It shows commerce MCP, which has spread mainly around product search and carts, starting to reach the stage where buyers and sellers reach agreement.
The next things to watch are the list of tools exposed to outside assistants and the approval rules that govern actions such as sending for signature. The real weight of this announcement will become clear when customers report numbers showing how deal lead times changed, beyond the convenience of reading data.
