Introducing Workspace MCP: agentic financial workflows, governed by design

AI inside investment firms has changed. A year ago, the conversation was about chat interfaces on top of financial data. Today, the firms moving fastest are running long-running agents, Claude Code, Codex, and similar systems that plan and execute work across hours without constant human input.

That shift has exposed a real problem. These agents pull from whatever data they can reach, often scattered across disconnected sources with no consistent permissions layer. Their outputs land in chat threads, documents, or someone's local folder, with no lineage, no access controls, and no clear path into production. The work gets done, but nobody can verify it, govern it, or build on top of it.

What MCP is

MCP, short for Model Context Protocol, is an open standard that lets AI agents connect to external data sources and take actions through them. Think of it as a universal adapter: instead of every agent needing a custom integration for every data source, MCP makes the connection standardized and plug-and-play.

Most companies have responded to this by shipping a read-only MCP integration. An agent connects, reads data, and answers a question. Done.

A read-only connection lets an agent answer questions. It doesn't let the agent do the job.

What Workspace MCP actually does

Workspace MCP turns OpenBB Workspace into an environment where MCP-compatible agents can do real financial work, not just retrieve data.

A connected agent can reach any data source you have set up in your workspace, whether that's a proprietary database, a licensed vendor feed, an internal system, or public market data. It can query that data, transform it, and combine sources that normally live far apart.

The output is reusable. Instead of a block of text that gets copied into a document and forgotten, an agent running through Workspace MCP produces working widgets, dashboards, and full apps that the rest of the team can open and use straight away. A portfolio manager's morning brief. A credit dashboard built overnight. A full application the whole desk keeps running.

Governed by design

When an analyst points a long-running agent at the workspace, everything that agent does runs through OpenBB Workspace's existing governance layer. There is no side door.

Entitlements apply to the agent exactly as they apply to the user behind it. A junior analyst's agent reads only what that analyst is permissioned for. It never sees the PM's positions.

The artifacts an agent builds are governed too. When an agent creates a widget, a dashboard, or a full app, that artifact inherits access controls like anything else in the workspace. An agent building something doesn't mean the whole firm can see it.

Connections are vaulted. API keys and credentials are managed centrally. An agent can use a data connection without the key ever being exposed or pasted into a prompt.

Every output carries data lineage. You can trace which sources fed into anything an agent produced. That matters when you need to confirm a licensed feed was used within its terms, or when compliance asks where a number came from.

Right now, a lot of firms are solving this the hard way. Analysts are shipping agent-built apps faster than anyone can review them, sending outputs to developers, asking colleagues how to get things into production. Some CTOs are hiring people whose main job is reviewing what the analysts' AI produced.

Workspace MCP removes that review step. The work is governed as it is created.

What you can do with Workspace MCP today

See it in action

Codex given a PRD and a data source. It builds the full Kalshi Event Explorer app from scratch: 10 widgets across 4 tabs, apps.json, widgets.json, and a live analysis of prediction market data.

Codex querying portfolio data, running a full risk assessment across multiple widgets, and pushing the output as a new widget into the Workspace.

Ihsan, our CPO, using Cortex Code inside OpenBB Workspace to write SQL queries against FactSet data on the fly, and pushing the output as a new widget into an existing app.

Full breakdown

Automate specific analysis workflows (eg. portfolio risk assessment)

Point an agent at your portfolio data and ask it to assess risk. In the second demo above, a single prompt to Codex, "assess the risk of my us_stocks portfolio according to the model calibration output and the factor covariance matrix," triggered the agent to read the Risk Calibration, Portfolio Risk, and Factor Covariance widgets in parallel. It came back with a full structured analysis: 19.09% total volatility, a 76.16% factor risk share, and specific covariance relationships flagging Energy, Liquidity, and Consumer Cyclical as the dominant risk sources. The analyst didn't write a line of code.

Build apps from your connected data

If your data is already connected to the Workspace, an agent can build a working application on top of it. An analyst describes what they need, the agent writes it, and the result is a functioning widget, a dashboard, or even a full app with multiple tabs, not a prototype that needs a developer to productionize. The firm's entitlements apply from the moment the artifact is created.

In the first demo above, Codex was given a PRD and access to Kalshi's public prediction market data. It built the full Kalshi Event Explorer from scratch: 10 widgets across 4 tabs covering event discovery, market inspection, orderbook data, and probability analysis, including the apps.json and widgets.json configuration files.

Ask an agent to explain what you're looking at

When an analyst opens a dashboard they didn't build, they can ask an agent to walk them through it. In the demo, Codex read the full Open Portfolio analytics app and explained each tab, its inputs, its configuration, and what the numbers meant, in plain language. That kind of institutional knowledge transfer normally takes a meeting. Here it takes 25 seconds.

Turn analysis into a shareable widget

After running an analysis, an agent can write the output directly into a new HTML widget inside your workspace. Both CoCo and Codex in the demos above created a new tab with a structured widget containing the full findings, ready to be shared with a portfolio manager or pinned to a dashboard. No copy-paste, no reformatting. The artifact lives in the Workspace, governed by the same access controls as everything else.

Long-running agents are going to run inside your firm whether you plan for it or not. Most of the output will end up in a chat thread, ungoverned, untraceable, and useful to nobody. Workspace MCP is how that changes.

Get started

Workspace MCP is hosted directly inside OpenBB Workspace. OpenBB provides the MCP endpoint from your Workspace session, and you authorize external agents with tokens created inside Workspace.

Connection steps:

Open OpenBB Workspace.
Click the hamburger menu in the top left and select MCP Companion.
Turn on the Endpoint connection. Workspace will show the hosted MCP endpoint for your session.
Click the "Create Token" button to create a token that will authorize your AI agent interacting with OpenBB Workspace through the MCP. Save the token when it is shown, because it cannot be viewed again later. You can create as many tokens as you need and they will work until you revoke them.
Configure your MCP client with the MCP endpoint and the authorization token. Here's an example how to configure the Claude Code to use the MCP:
```
claude mcp add --transport http workspace_mcp \
  https://pro.openbb.co/mcp \
  --header "Authorization: Bearer obb_mcp_..."
```
Keep the Workspace tab open and the MCP Companion connected while the agent works.

Once connected, the agent can read your active Workspace session, manage widgets, build apps from your connected data, and navigate dashboards on your behalf. The agent runs through the same Workspace governance layer as the user behind the session: entitlements still apply, credentials stay vaulted, and artifacts created by the agent remain governed inside Workspace.

For the full setup guide, see the Workspace MCP Quickstart. For architecture and security details, see the Workspace MCP Overview.

For teams that want to run the full setup inside their own premises or VPC, reach out to our team, and we can walk through what that looks like.

TL;DR: We are open-sourcing the entire OpenBB product suite under a permissive open source license

Today is bittersweet.

On December 20, 2020, over the Christmas holidays, I wrote the first lines of Gamestonk Terminal, what would eventually become OpenBB. My flight home to visit my parents had been cancelled because of COVID, so I stayed in London and started building a tool to streamline my own investment research process.

At the time, the idea was simple: individuals (and firms) should be able to own their research platforms. They shouldn't have to adapt their workflows to whatever a data or software vendor decided to build. They should own the entire experience - from the data they connect to, to the interface analysts and PMs use every day, to the APIs, models, skills, tools and AI agents that increasingly form part of the investment process.

Over the last almost six years, OpenBB evolved far beyond anything I imagined when writing those first lines of code. We started as an open-source terminal and went on to build the SDK (now Open Data Platform), the OpenBB Bot, the OpenBB Workspace, the OpenBB Copilot, the Excel Add-in, and an ecosystem of applications created by our team, our partners and the community.

We kept innovating and being at the forefront of what user experience should be. I was prepared to die on the hill that if we were to become the financial infrastructure software for the buy-side and sell-side, then we could not monetize data. Monetizing data would have made us a data vendor. The margins would have been higher, but the incentive would have shifted from offering a better UI/UX to selling more datasets. Selling an infrastructure platform is incredibly challenging, for many reasons. And so we died on that hill. Along the way, we built an incredible community, reached millions of people through our open-source project, worked with some of the largest financial institutions in the world and assembled a team that consistently built far beyond what should have been possible for a company of 10 people. There is a lot to be proud of.

But despite all of that, we couldn't find the product-market fit needed to build a sustainable business around this vision within the time we had.

As a founder, I've always bet the house on the next customer, the next feature or the next launch to change the trajectory of the company. Until even just last weeks, when we announced self-serve Workspace Lite. If we were ever going to close doors, then we never wanted to look back and think "what if".

In hindsight, we could obviously have made different decisions - e.g., surrounding data, or going more vertical with clients and their workflows. But short-term monetization was not something I was prioritizing over where I thought the industry was heading. Ultimately, I wanted us to stand for something.

In the last phase of the company, I explored many paths to give OpenBB a better home. We spoke with larger companies that shared parts of our vision and tried to find a home where the products, and ideally the team behind them, could continue to grow. Ultimately, we weren't able to make that happen.

I left London with my wife to build OpenBB and dedicated almost six years of my life to it. So did the team. We built at the intersection of finance, AI, open source and software infrastructure, often working on problems before they became obvious to the broader market.

Before MCP existed, we had created our own API protocol so agents could interact with the data and analytics widgets in the workspace. In 2023, we built askobb, which let anyone ask investment research questions in natural language that required joining multiple different datasets together.

What this team created deserves to continue existing.

More importantly, I still believe the original vision is inevitable. The future of financial software is not a single platform every firm is forced to use. It is thousands of firms building environments that reflect how they actually work - their own data, internal systems, investment processes, risk models and compliance requirements. Increasingly, their own APIs, MCP servers, models and AI agents too.

If that future is coming, then the technology we built shouldn't disappear simply because we weren't able to commercialize it successfully.

It should become available to everyone.

Today, with the support of the team and OSS Capital, we are committing to releasing the entire OpenBB product suite under a permissive license. This includes OpenBB Workspace, Open Data Platform, OpenBB Copilot and the OpenBB Excel Add-in.

These products represent over 5 years of engineering, millions of dollars invested in R&D and thousands of decisions, experiments and iterations with users. They will become a foundation that individuals, startups, data providers and financial institutions can freely use, modify and build on top of.

We will share more details about the order and timing of each release as we complete that work. In parallel, we will determine the right long-term structure to steward the projects, support contributors and preserve what made OpenBB special in the first place. Existing customers and users of the hosted products will hear from us directly about timelines.

For the partners who built applications for the OpenBB ecosystem, I hope this decision makes your products even more valuable. You already did the work of turning your datasets and analytics into applications that users can interact with. Now, those applications will be able to run inside infrastructure that firms can fully own, extend and customize, while combining them with data and tools from other providers across the ecosystem.

The same applies to the broader community. Developers will be able to use the entire OpenBB stack as a starting point rather than rebuilding the same infrastructure from scratch. Firms will be able to deploy it, adapt it to their requirements and connect it to the systems where their differentiated knowledge already lives.

Over the years, many talented people helped make OpenBB what it is today - employees, contributors, partners. Every one of them left a mark on the product.

But I want to recognize the people who carried OpenBB to the very end. Through the uncertainty and the final stretch, they kept building. They are engineers, product builders, designers and operators who have worked across financial data, AI, developer infrastructure and open source. In alphabetical order, they are:

Andrew Kenreich, Head of Product Engineering - LinkedIn
Darren Lee, Software Engineer - GitHub, LinkedIn
Ihsan Saracgil, CPO - LinkedIn
José Donato, Software Engineer - LinkedIn, X, GitHub, Website
Juan Alfonso, Software Engineer - GitHub, LinkedIn
Minh Hoang, Head of Product - LinkedIn, GitHub
Ogonna Nnamani, DevOps - LinkedIn, Medium
Rita Figueiredo, Head of Marketing - LinkedIn, Website
Rita Soares, Head of Design - LinkedIn, Website
Theodore Aptekarev, CTO - LinkedIn, GitHub

To our customers: thank you for trusting a small team with such an ambitious vision.

To our partners: thank you for building alongside us and helping create a more open financial data ecosystem. I hope the next chapter gives you even more freedom to serve your users.

To our investors: thank you for believing in us, including when OpenBB was little more than an idea being built from my living room in London. In particular, I want to thank OSS Capital and Joseph Jacks for supporting this decision and enabling the technology to live beyond the company. Two people I want to name separately: Justin Hoffman and Larry Augustin - working with both of you made me a better founder, but more importantly, a better person.

To every contributor who opened a pull request, reported a bug, wrote documentation, answered a question in Discord, built an integration or simply told someone else about OpenBB: thank you. OpenBB would not have been possible without you.

Finally, to every person who spent part of their career building OpenBB: thank you. We pushed the industry forward and proved that world-class financial infrastructure can be built in the open. The commercial outcome doesn't change the quality of the work.

OpenBB didn't become the company I imagined when I started this journey. But the mission was always larger than the company, and I still believe the ideas behind it are right.

If, ten years from now, firms around the world are using OpenBB as the foundation for software they truly own - connecting their own data, building their own workflows and deploying their own AI agents - then what we built will have achieved something that lasts far beyond us; which was my goal all along: have an impact.

Thank you for one hell of a ride.

Didier Lopes
(LinkedIn, X, GitHub)