Ignition MCP server: what it does and how to connect it

Most of your software lives in its own world. You log in, do the work, log out. Change tabs, context switch, lose your place, get distracted and eventually remember where you left off.

If you want information from another tool, you copy and paste it across. If you want two platforms to talk to each other, you wait for them to build an integration, pay for a Zapier connection, or do it by hand.

That model is rapidly changing thanks to AI. And the Ignition MCP is one of the clearest examples yet of what comes next.

The Ignition MCP server is Ignition’s official connection for AI assistants. Once it is connected, Claude, ChatGPT or another MCP-compatible assistant can look things up and make changes in your Ignition account when you ask in plain English, instead of you clicking through the app. Ignition announced it at AICPA ENGAGE in June 2026.

According to Ignition’s help centre (as at September 2026, last updated 20 August 2026), it covers:

• What it reaches: clients and contacts, proposals, deals, invoices, payments and billing, services, templates and forms. Some actions are read-only lookups; others make changes on your behalf.
• What it costs: it is available on every Ignition plan at no extra cost. If your plan does not include a feature such as Deals or Forms, the matching actions are not available through the MCP either.
• What it is allowed to do: the assistant inherits the permissions of your Ignition role, so it can only do what your own login can do, and only inside the practice you are signed in to.
• Where it stops: invoices it creates are drafts. Someone still confirms them in Ignition before they are issued.

MCP stands for Model Context Protocol. It is an open standard, introduced by Anthropic, that lets AI assistants connect directly to software platforms.

Think of it this way. Claude, or any AI assistant built on MCP, can read from and write to a platform in real time using natural language. No copy-paste. No switching tabs. No exporting data so the AI can see it.

When a platform builds an MCP server, it is essentially giving Claude a direct line into its data and functionality. Claude can ask Ignition what proposals exist, what services are in the library, what a client’s status is, and then act on that information, all without the user touching the Ignition interface.

The MCP standard is open, which means any software company can build one. Ignition has. If you want the longer, jargon-free version of how this works across a firm’s whole stack, start with our explainer on MCP for accountants.

Ignition’s own instructions are three steps, and they are the same whichever assistant you use:

1. In your AI app, add a new MCP connection and enter the Ignition server address: https://mcp.ignitionapp.com/mcp
2. You are taken to Ignition to sign in and approve access.
3. Once approved, you are connected and can start asking your assistant to work in Ignition.

Where you add that connection depends on the assistant. In Claude, per Anthropic’s help centre as at September 2026, individual Pro and Max users go to Customize, then Connectors, click “+” and choose “Add custom connector”, then paste the address. On Team and Enterprise plans only an Owner can add a custom connector (under Organization settings, then Connectors); members then switch it on for themselves. Free Claude accounts are limited to one custom connector. For ChatGPT and other apps, the steps vary by app and plan, so follow Ignition’s help article and your assistant’s own documentation rather than a screenshot that may already be out of date.

To disconnect, go to Profile, then Connected apps in Ignition, and remove the access in your AI app as well.

One decision to make before anyone clicks “approve”: because the assistant acts with the connecting person’s permissions, whoever connects it decides what it can touch. Settle who may connect, and from which login, in your one-page AI policy first.

Proposals are time-intensive because they are context-intensive. A good proposal requires knowing what was discussed in the discovery call, what the client’s situation is, how similar engagements have been scoped and priced, and how to write an intro that actually converts.

That knowledge typically lives across three or four places: a meeting recording, a CRM, a previous proposal, and someone’s memory. Pulling it together manually is exactly the kind of work that slows firms down and creates inconsistency in how proposals get written and sent.

This is the problem the Ignition MCP, paired with an AI meeting tool like Vinyl, is designed to dissolve.

The session that demonstrated this started with a discovery call transcript and ended with two completed, personalised draft proposals in Ignition. The time taken was under four minutes. (The full walkthrough, step by step, is in I built two proposals in under four minutes using MCP.)

Here is what Claude actually did across that workflow.

Read the practice data. Claude pulled the practice information directly from Ignition, listed all 20 recent proposals and their statuses, and catalogued the full service library, more than 50 services, without any manual navigation.

Analysed the discovery call. Using Vinyl, Claude read the full transcript from the discovery call. It identified the scope, the client’s situation, pricing signals, and campaign angles, then mapped that conversation to a concrete proposal structure.

Handled the client and contact work. Claude looked up the relevant client, confirmed the email address was already in the system, and resolved a duplicate record to identify the correct active client to use.

Built the proposals. Two drafts were created directly in Ignition. The primary proposal was structured as a GTM Pilot Campaign with a three-month minimum, auto-expiry turned off, the correct email template and terms attached, and a fully personalised intro message written and loaded in.

Applied pricing intelligence. Claude cross-referenced a lost proposal and an accepted proposal to inform its recommendation. It narrowed the scope to better fit the pilot framing and end-of-financial-year timing, and flagged a missing service, a Marketing List, that the original scope had overlooked.

All of that happened through a conversation with Claude. No manual data entry. No tab switching. No reformatting.

When this session was recorded, the Ignition MCP was in private beta. As at September 2026 Ignition’s help centre lists it on every plan, so the question has moved from “can we get it?” to “what should we point it at?”. Three things are worth understanding before you connect it.

The quality of the output depends on the quality of the input. Claude can only work with what it can read. If the discovery call was captured clearly by a tool like Vinyl, the proposal will reflect that. If the discovery call was vague, or never recorded, the AI has nothing to work from. Structured discovery conversations become more valuable, not less, as this kind of tooling matures.

The MCP layer is broader than Ignition. As more platforms build MCP servers (practice management systems, CRMs, document tools) the manual handoffs between them become optional. Proposal software, billing, and client management can start operating as a connected environment rather than a collection of separate logins. Ignition is one piece of that picture, and our AI for accounting firms hub tracks the rest.

The human role shifts, not disappears. What Claude produced in this session were drafts. A practitioner still reviewed the pricing rationale, confirmed the scope, and would send the proposal. The difference is that the groundwork, the reading, the cross-referencing, the writing, was done in minutes rather than the better part of an hour.

The Ignition MCP is one of the first purpose-built integrations of this kind for accounting and bookkeeping software. It will not be the last.

Firms that understand how MCPs work, and start building workflows around them, will have a structural advantage as more platforms open up this layer. The concept is straightforward: the AI reads your tools, reasons across them, and acts on your behalf. The hard part is knowing which problems are worth pointing it at.

Proposal creation, as this session showed, is a strong place to start.

So here is the question worth sitting with: if the research, the cross-referencing, the writing, and the data entry can all happen in under four minutes, what is the real cost of doing it manually?

Does the Ignition MCP server cost extra?

Not as at September 2026: Ignition says it is included on every plan at no extra cost. What your plan does change is reach. The MCP can only use features your subscription already has, so a firm without Deals or Forms will find those actions missing. Your AI assistant is a separate subscription, and on Claude’s free tier you only get one custom connector, which matters if you want Ignition alongside another tool.

Can the Ignition MCP send a proposal or issue an invoice without me?

It acts with the permissions of whoever connected it, so the honest answer is “whatever that login can do”. Invoices are the stated exception: Ignition says they are created as drafts that someone confirms in the app. For everything else, treat the assistant’s output as a draft you review, and connect it from a login whose permissions you would be comfortable handing to a new starter.

Is MCP for accountants only about Ignition?

No. MCP is an open standard, so any vendor can publish a server, and more of a firm’s stack will offer one over time. Ignition is useful as a first test because proposals sit where a lot of firm knowledge meets, but the same connect, approve and ask pattern applies to any MCP server. Check each vendor’s own documentation for what it can read and change, because coverage varies widely.

Why are proposals a good first use case?

Because proposal work is context-intensive rather than difficult. A good proposal needs what was said on the discovery call, the client situation, how similar work has been scoped and priced, and an intro that converts. That knowledge normally sits across a recording, a CRM, an old proposal and somebody’s memory, which is exactly the kind of assembly work that slows firms down and makes proposals inconsistent.

How long does the workflow take?

The demonstrated session went from a discovery call transcript to two completed, personalised draft proposals in Ignition in under four minutes. Treat that as what the workflow can do in a clean run, not as a guaranteed time on a messy file. Your first run will be slower, because you will be checking what the assistant can see.