Instrumentl, the grant management platform used by non-profits to find and track funding, has put its first public machine interface into the world, and it is an MCP server rather than a REST API. In Instrumentl MCP server, a help article by Nate White updated this week, the pitch is direct: “Connect your AI assistant to Instrumentl and take real action on your grant data — directly from a conversation.” The server exposes 25 tools, 13 that read, account, users, projects, matches, tracked grants, funder profiles, budgets, expenses, and 11 that write, saving matches, updating tracked grants, creating tasks, logging expenses. Claude and ChatGPT connect through OAuth sign-in, with Copilot and Gemini on the roadmap.
There are no benchmark numbers, only product rules, and the rules are the interesting part. The server is available to every paid plan and every active trial, but trial accounts “are limited to 15 records returned per project per session” on the matching and search tools, which is a quota expressed as a tool behavior rather than a rate limit. Access follows the product: “Your AI assistant operates under the same governance model as the rest of Instrumentl: workspace-level access controls and user permissions.” The promise is the one every connector makes, “you can describe what you want in a conversation and your AI handles the rest,” and the write tools are what make it more than a promise. An assistant can change what a grant is worth and what has been spent against it.
The catalog’s record of Instrumentl is as thin as the surface it describes, which is the story. The Instrumentl provider page lists one API page, the Instrumentl MCP Server, and nothing else. On the Agent Readiness score the MCP server is lit, and so is the whole identity layer, delegated identity, protected resource metadata, and dynamic client registration, which matches an OAuth connector that works from the Claude and ChatGPT directories. The spec presence dimension is unlit, because there is no contract to find. Agentic access is unlit for the same reason: the catalog cannot say which of the 25 tools act and which only read, because the help article is the only description of them.
The Kin Score is 25.2, emerging band. Discoverability carries it at 75.0 and access clarity at 44.7. Contract quality is 0.0 and contract governance is 0.0. Developer ergonomics is 19.0 and operational transparency 15.8. The Agent Readiness score is 27.3, agent-aware, higher than the composite would suggest, because the agent door is registered and the developer door does not exist. Instrumentl is a clean example of a company whose first API is an MCP server. Eleven of its tools write to grant records and budgets, and there is no machine-readable contract that says what they accept or return. The conversation handles the rest. The catalog cannot yet read what the rest is.