Impala MCP Server Brings Philanthropy Data to Claude and ChatGPT

What Impala launched

Impala is a New York data company that sells intelligence about the philanthropic sector. On August 19, 2026, it released a Model Context Protocol (MCP) server that lets Claude and ChatGPT query its database of funders, nonprofits and grants directly.1 The company describes the server as an intelligence layer. Foundations, grantmakers and nonprofits can use it to build their own AI applications and workflows on top of Impala's data, instead of working only inside Impala's web dashboards.1

The data covers funders, nonprofits, grant histories, governance, financial health and ecosystem relationships across the United States.1 Impala's own website gives the scale as more than 250,000 grantmakers, more than 2 million nonprofit profiles, more than 25 million grants and about 10 million board members.10 The company says users can ask questions in plain language and get answers drawn from what it calls IRS-verified philanthropic data. They can also compare their funding strategy against peer funders and wider sector trends.1

The server works now in Claude and ChatGPT, and Impala says support for more AI environments will follow.1 It is also listed in Anthropic's Connectors Directory, which the company notes came after Anthropic reviewed it for security, reliability and compatibility.1

The MCP angle: a niche data company moves into the assistant

The interesting part is less about philanthropy than about where specialized data is now being delivered. Anthropic introduced MCP in November 2024 as an open standard for connecting AI assistants to outside data and tools. It was meant to solve the "N×M" problem, where every assistant needed a custom connector for every data source.13 In December 2025, Anthropic donated the protocol to the Agentic AI Foundation, a fund under the Linux Foundation co-founded with Block and OpenAI.11 At that point Anthropic counted more than 10,000 active public MCP servers and more than 97 million monthly SDK downloads. It also listed ChatGPT, Gemini, Microsoft Copilot, Cursor and Visual Studio Code among the products that support the protocol.15

Those numbers explain why a fairly small vertical data provider would build a server rather than a dedicated plugin for each assistant. One open-data nonprofit that recently added MCP support made the same point about its own work. It wrote a single integration, and that one integration reached as many AI chat interfaces as it could realistically support.14 Impala is making the same bet. It ships one server, launches on the two largest consumer assistants, and plans to add others later.1

One industry guide calls this a recognizable business model: proprietary data sold behind a paid license, with MCP serving only as the delivery method. It names Bloomberg, CB Insights and Lex Machina as examples.19 That guide's own estimate puts commercial MCP revenue at around $400 million a year as of May 2026. Treat that figure as directional, not authoritative.19 Even so, Impala fits the pattern. Its main asset is a curated dataset about relationships between funders and grantees, and MCP lets that dataset follow users into the tools they already open every day.

The Jewish Federation Bay Area is the main customer quoted at launch. Its data operations chief said the practical benefit is no longer switching between platforms to answer questions specific to the organization.2 Every MCP vendor makes this pitch. In this case, though, it addresses a real problem in a sector where grant management systems, donor CRMs and research databases often don't connect to each other.

What users are supposed to build

Impala's chief product officer, Marisa Sires, gave three examples of applications the company expects: a board-reporting tool built around a foundation's grantees, a due-diligence agent tuned to an investment thesis, and a portfolio-analysis workflow matched to a particular strategy.2 The company also says customers can combine its data with their own grant management systems, fundraising CRMs and internal records. It specifically mentions Claude Code and OpenAI Codex as tools for building custom dashboards and decision aids.1 Other listed uses include mapping whole funding ecosystems by geography and cause area, and finding potential funders whose priorities match a nonprofit's work.8

Impala also offers paid services to build custom agents and applications for organizations that don't want to do it themselves.1 The release says Impala is the first company to pair sector-wide philanthropic data delivered over MCP with this kind of build service.2 That claim comes only from the company, and nothing in the coverage tests it.

The services offer is probably the most commercially important part of the launch. Fast Forward's 2026 AI for Humanity Report surveyed AI-focused nonprofits and found that 26% have no technical staff. Nearly every leader interviewed said finding and paying for AI talent was a barrier.16 A Center for Effective Philanthropy study from 2025 found that almost two-thirds of foundations and nonprofits said none or only a few of their staff had a solid understanding of AI.20 Pointing these organizations to Claude Code won't help much if no one on staff can use it. The real market may be Impala building the tools for them, with the MCP server as the base layer.

Where the coverage agrees and where it falls short

The reporting is unusually uniform. Nearly every outlet that covered the launch, from local newspaper sites to financial terminals, ran the same distributed press release.24679 Dealroom's short summary was one of the few rewrites, and it repeated the same points.5 So the coverage agrees on every fact, but almost none of it is independent. None of the reporting tests how accurate the data is when it reaches an assistant, how well the tools work, or how the server compares with competitors.

The one real discrepancy is about pricing. The release tells interested organizations to contact Impala, which suggests a sales process.1 Impala's homepage, by contrast, says the server is included with a customer's plan.10 Both can be true: existing subscribers may get it automatically while new customers go through sales. Still, the company hasn't clearly explained who can use the server and on what terms. That matters because the homepage also advertises free access to basic profiles.10

The risks are mostly about deployment

When a database of nonprofit finances, board rosters and grant flows is connected to general-purpose AI assistants, questions follow about accuracy and data handling. Those concerns are already present in the sector. In Fast Forward's survey, 88% of AI-focused nonprofits worried about giving out inaccurate information and 84% worried about exposing or misusing personal data.16 Writing about MCP more broadly, the Data Transfer Initiative noted that connecting agents to data can add AI providers as another party with access. It also warned that prompt-injection attacks can lead agents to mishandle what they retrieve.14 Prompt injection, where malicious text in a data source instructs a model to misuse other tools, is described as the most commonly reported type of MCP incident. Mitigations depend on how a system is deployed more than on the protocol itself.19

Most of Impala's data comes from public filings, so the privacy risk is lower than in health care or consumer finance. The bigger risk for a foundation is a model misreading a 990 filing or inventing a grant relationship in a board memo. Impala's emphasis on IRS-verified data is aimed at that problem.1 Anthropic's directory review also offers some assurance about security and reliability.1 Neither guarantees that answers will be correct.

Why it matters

Taken alone, this is a modest product launch from a niche vendor. As a signal, it shows how much MCP has changed in less than two years: it began as an experiment for developers and is now standard infrastructure for specialized data companies. Public-sector work shows the same trend. The U.S. General Services Administration is running a fall 2026 hackathon to prototype MCP servers for federal open data, betting that plain-language AI access will make government information easier to use.18 The protocol is also still changing. A July 2026 revision of the specification made MCP stateless at the protocol layer, which maintainers called the biggest change since authorization was added.13

Philanthropy usually adopts new technology slowly, and funders have done little to pay for grantees' technology. Fast Forward found that 78% of AI-focused nonprofits wish funders would pay for long-term infrastructure instead of short pilots.16 Impala's launch puts a common data layer in the place where nonprofit and foundation staff increasingly do their work. Whether it succeeds will depend less on the protocol than on two things: whether the data holds up under the questions people ask, and whether organizations without engineers end up paying Impala to build tools they can't build themselves.