Your System of Record plus MCP is the Secret to Useful Institutional AI

In our recent article on The Investment System of Record, we made a claim that deserved its own treatment: that the System of Record (SoR) is not just a passive beneficiary of AI, but the active foundation that makes AI useful – and that AI, in turn, makes the SoR more valuable. This is the article about how those two directions work in practice using an MCP Server.

The investment industry has spent the past two years focused on what AI can read. Summarise this filing. Transcribe that call. Extract guidance from this slide deck. These are genuine capabilities but they solve only half the problem.

The other half is about what happens to the output. Where does the AI’s work live? Who can see it? Is it linked to the company, the thesis, the decision it informed? Can you find it in three years when the team wants to review what the thesis looked like at the time of a specific trade?

If the answer is “it’s in a chat history somewhere,” you have built a faster way to lose institutional knowledge.

.

.

The Two Directions of Flow

A connected AI workflow has two directions, and both matter equally.

Direction one: SoR → AI. Your System of Record feeds context into whatever AI tooling your team uses. When an analyst asks an AI agent to prepare a pre-meeting briefing, the agent doesn’t start from a blank slate of public filings. It queries CalibreRMS via API or Model Context Protocol (MCP) and retrieves the firm’s prior meeting notes, the current investment thesis and its milestones, the latest scorecard assessments, the engagement history, the model forecasts, and the counter-thesis. The result is a briefing grounded in your firm’s own research, structured around what your process says matters.

This is what MCP connectivity makes possible. The Model Context Protocol provides a standardised way for AI tools – ChatGPT, Claude, custom agents, internal copilots – to discover and query your SoR as a live data source. Your analyst working in their preferred AI environment can pull thesis milestones, scorecard histories, or engagement records directly into the conversation, without copy-pasting, without context switching, and without exposing data to unsanctioned channels. The SoR becomes the institutional memory that AI interactions can draw from.

When the analyst is using AI chat for exploratory research, in addition to CalibreRMS the AI workflow may pull data from the MCP of other systems such as Factset or Sustainalytics or Visible Alpha as well as web searches, deep research and bespoke MCP connectors the team has set up to allow human analysts to reason over all relevant information.

Direction two: AI → SoR. The AI’s finished work flows back into CalibreRMS as a permanent, structured, auditable record. Not the raw chat transcript. Not the intermediate reasoning. The final, reviewed output – the deep research note, the scorecard assessment, the thesis evaluation – saved as a first-class artifact, timestamped, tagged to the relevant company, linked to the thesis it supports, and visible to the entire team.

When sending research back into your SoR, the last thing you want polluting your system is AI slop. Clear structures for information classification and publication statuses is a linchpin for AI generated thesis monitoring and development. The professional investor human-in-the-loop is the highest form of quality control.

This second direction is where most firms fall down. They deploy impressive AI capabilities but leave the outputs stranded in personal workflows or fragmented across multiple unconnected systems. The PM never sees them. Compliance cannot audit them. The next analyst to cover the name starts from scratch. The AI did some useful work but the institution lost it.

.

.

Saving AI Research as Permanent Records

Consider a concrete workflow. An analyst uses a deep research agent – whether CalibreRMS Intelligence, Claude with MCP access, or a proprietary internal tool – to produce a comprehensive analysis of a company’s capital allocation over the past decade. The agent reviews annual reports, models cash flow deployment, scores acquisition discipline, and produces a structured assessment with citations back to source documents.

That output is valuable. But its value is zero to the institution if it lives in a chat window.

In a connected workflow, the analyst reviews the AI-generated analysis, applies their judgement – agreeing with some conclusions, overriding others, adding context the model could not know – and publishes the final version back into CalibreRMS as a structured research note via the MCP connection. The note carries clear provenance: AI-assisted, human-reviewed. It is tagged to the company record, linked to the investment thesis, and timestamped in the time-series database. It sits alongside every other piece of research the team has produced on that name – the hand-written meeting notes from two years ago, the Excel model, the ESG engagement record, the proxy voting rationale.

Now the PM can see it. The next analyst can build on it. The compliance team can audit it. The consultant reviewing your process in three years can verify that the research existed before the position was initiated. The AI assisted, human crafted output has become institutional memory.

This is also how firms solve the normalisation problem we raised in The Goal is Better Decisions. When every analyst has their own agent producing research at unprecedented volume, the risk is fragmentation. By routing final AI outputs through the SoR, firms ensure that AI-generated research meets the same structural standards as human-authored work: same templates, same tagging, same audit trail. The RMS remains the single decision plane where everything converges.

MCP: Your SoR as a Live Data Source for Any AI Tool

The CalibreRMS MCP server exposes the platform’s data model to any MCP-compatible AI client. In practical terms, this means an analyst working in Claude Desktop, a custom-built research agent, or any tool that speaks MCP can query CalibreRMS without leaving their workflow.

The queries are not generic document retrieval. Because CalibreRMS structures data against the investment lifecycle – thesis milestones, scorecard frameworks, engagement objectives, decision logs – the AI tool receives semantically rich context. It does not get a pile of documents to summarise. It gets the firm’s rulebook: the priority placed on specific scorecards, the stages of the investment lifecycle, the evidence required to move from watchlist to conviction. This is what transforms a general-purpose AI into an extension of your team’s specific investment philosophy.

Critically, MCP access inherits the same permission model and security controls as the rest of CalibreRMS. Role-based access ensures an AI agent sees only what the requesting user is authorised to see. Every query is logged. Data never leaves the firm’s security boundary. The Bring-Your-Own-LLM architecture means the AI model processing the query runs inside the firm’s own cloud tenancy – the SoR provides context, but the intelligence engine remains under your governance.

The Flywheel: Why This Gets Better Over Time

The connected workflow creates a compounding loop. Every AI-assisted, human verified note saved back into the SoR enriches the proprietary corpus. The richer the corpus, the more valuable the next AI query becomes – because the agent has deeper context to draw from. A pre-meeting briefing prepared against five years of structured engagement history, scorecard trends, and thesis evolution is qualitatively different from one prepared against public filings alone.

This is the flywheel that converts AI from a productivity tool into a durable competitive asset. And it is only possible when the SoR is both the source of context and the destination for output.

The firms building this loop today are not necessarily the ones with the most sophisticated AI models. They are the ones whose institutional plumbing – their System of Record – is open enough to feed AI and structured enough to absorb its output. They treat AI providers as interchangeable intelligence engines and their SoR as the constant.

The Practical Starting Point

For teams ready to connect these workflows, the path is simpler than it might appear. CalibreRMS already provides both the API and MCP server access required. The steps are:

Connect your AI tools to CalibreRMS via MCP so they can query your proprietary research, theses, internal financial forecasts, valuations and scorecards as live context.

Establish a “save back” discipline: when an AI agent produces a final research output, the analyst reviews it and publishes it into CalibreRMS as a permanent record, tagged and linked like any human-authored note.

Mark provenance clearly: AI-assisted outputs carry that designation so the team and compliance can distinguish them from purely human work.

Let the corpus compound: every saved artifact makes the next AI query smarter.

The SoR was always designed to be the institutional memory of the firm. AI is simply the most powerful tool yet for both querying that memory and enriching it. The firms that connect these two directions – SoR powering AI, AI feeding the SoR – will build a research capability that appreciates with every interaction.

.

Interactive Demo