MCPFlo is a client for the Model Context Protocol (MCP) — an open protocol that lets an application expose capabilities to an AI model (or, in MCPFlo’s case, to you directly) in a structured, discoverable way. If you’re new to MCP, these are the primitives you’ll see throughout the app.

Tools

A tool is an action the server can perform — the MCP equivalent of a function call. Each tool declares a JSON Schema for its input, so a client knows what arguments to send before calling it. In MCPFlo, tools render as an auto-generated form based on that schema; you fill it in, invoke the tool, and see the raw result.

Example: a search_files tool might take a query string and a path, and return a list of matches.

Resources

A resource is a piece of data the server can hand over — a file, a database record, a URL’s contents — identified by a URI. Unlike a tool, reading a resource has no side effects; it’s just fetching data. Resources can be text or binary (MCPFlo renders both), and a server can also link to a resource from within a tool’s response.

Example: a resource might expose file:///project/README.md or postgres://db/table/rows.

Prompts

A prompt is a reusable, parameterized template for a conversation — a predefined set of message turns the server suggests, which can take named arguments. Rendering a prompt in MCPFlo expands it into the actual message turns it produces, so you can see exactly what the server would hand to a model.

Example: a code_review prompt might take a diff argument and expand into a system + user message pair ready to send to an LLM.

Elicitation

Elicitation (elicitation/create) lets a server ask the user for more information mid-call — for example, a tool that needs a missing parameter or a confirmation before proceeding. When a server elicits input, MCPFlo renders the requested schema as a form right there in the call flow, and sends your answer back to resume the tool. (Only form-mode elicitation is supported; URL-mode is not.)

Sampling

Sampling (sampling/createMessage) is the reverse relationship: it lets a server ask the client to run an LLM completion on its behalf — useful for servers that need a model’s help to finish a task. Since MCPFlo has no model in the loop, it surfaces the sampling request to you and lets you type the response by hand. This keeps every interaction deterministic and token-free, and lets you see exactly what a server is asking a model to do.

How they fit together

A single MCP server can expose any mix of tools, resources, and prompts. When you connect to a server in MCPFlo, it performs capability discovery — asking the server what it offers — and builds the capability tree you browse in the sidebar. From there, invoking a tool, reading a resource, or rendering a prompt all follow the same request/response pattern over the connection’s transport (stdio or Streamable HTTP).