CalderFlow MCP¶
A process that lives in a paragraph cannot be reviewed. The moment a workflow has to be argued about, it needs a shape: a flowchart for a sequence with branches, a funnel for stages that lose people, a lifecycle for phases and milestones, a timeline for events, an entity diagram for tables and their keys. CalderFlow turns a description into one of those five shapes and returns a share link, which is the difference between a diagram you describe and a diagram a colleague can open.
The server is worth a page for two reasons that are separate from the product. It is the latest clear example of an MCP App: its resources/list serves a ui://calderflow/chart-<id> resource typed text/html, so the chart renders inside the conversation instead of arriving as a link a reader has to trust. And its evidence tiers are split cleanly, with the handshake and the full schema open to anyone and only tools/call behind a key.
Server type: remote Streamable HTTP
Endpoint: https://www.calderflow.com/mcp
Server: calderflow, version 0.1.0, title "CalderFlow"
Protocol: 2025-06-18
Auth: API key for tool calls; initialize, tools/list and resources/list are open
Tools: 10
Resources: ui://calderflow/chart-<id>, text/html (inline chart viewer)
Prompts: none (prompts/list returns -32601)
Listed on: mcp.so feed, stamped 2026-11-17T04:43:17Z
The evidence tier¶
Handshake open, calls gated. initialize answered 200 as calderflow 0.1.0 on protocol 2025-06-18, declaring tools and resources capabilities, a website URL and two icon assets. tools/list returned ten closed schemas with their required arguments. resources/list returned the chart viewer resource, which is the MCP Apps pattern rather than a text-only answer. prompts/list returned JSON-RPC -32601, Method not found, so there is no prompt surface to enumerate.
Then tools/call to create_flowchart answered HTTP 200 with isError: true and one sentence: Authentication required: send 'x-api-key:
Tools & Capabilities¶
| Tool | Required arguments | What it produces |
|---|---|---|
create_flowchart |
title, mermaid |
A flowchart from Mermaid source |
create_chevron_chart |
title, steps |
A chevron process strip |
create_funnel_chart |
title, stages |
A funnel with stage values |
create_lifecycle_chart |
title, phases, milestones |
Phases with milestones on each |
create_timeline_chart |
title, events |
An event timeline |
create_erd_chart |
title, entities |
An entity diagram with entities and their fields |
edit_chart |
chart_id |
Edits an existing chart |
get_chart |
chart_id |
Reads one chart back |
list_charts |
none | The charts in the account |
delete_chart |
chart_id |
Removes a chart |
Installation¶
claude mcp add calderflow --transport http https://www.calderflow.com/mcp
An API key can travel in an x-api-key header, as a ?key= query parameter, or as a bearer token. The MCP App resource is what makes the chart appear inline in a client that supports MCP Apps; clients without that support can still be given the returned share link.
Why This Matters for Operators¶
The cost of a process diagram is not the drawing, it is the retyping: someone explains the order-to-cash flow in a chat, someone else redraws it in a slide tool, and the two versions diverge. Describing the process once and getting an editable chart back removes that second pass, and an <iframe>-renderable HTML resource means the artefact can be reviewed in the same thread where the process was described.
Questions operators ask¶
Why does an open schema matter if the calls are gated?¶
Because it settles what the tool can and cannot do without a sales conversation. The ten schemas name five chart families, one edit path, one read path and a delete; there is no data import and no third-party source. A team can decide whether that shape fits before it asks for a key.
Is the chart data mine or the vendor's?¶
The chart lives on the vendor's side: every tool returns a chart_id and the viewer is a hosted resource. delete_chart exists, so removal is in the surface rather than in a support ticket.
Does it work in any MCP client?¶
The transport and protocol version are the standard ones, so any Streamable HTTP client can call it. The inline chart, however, needs a client that renders MCP App resources; elsewhere the output is the share link.
Related¶
- MCP Apps - Interactive UI in the Client - how the
ui://resource pattern works - MCP Specification 2026-07-28 - the protocol revision this server's 2025-06-18 handshake sits behind
- Ryze Google Ads MCP - an example of a hosted reporting surface that renders tables rather than diagrams
Integration with CorpusIQ¶
CorpusIQ is the connector layer that lets AI assistants read the systems a business already runs on - accounting, payments, CRM, email, files and analytics - and answer with the source named. It is read-only by design: it retrieves data and never writes to, or acts in, a connected system, which is what makes it safe to point at live accounts.
This page is part of the CorpusIQ directory, the reference we maintain for the MCP server and agent-skill ecosystem. Run alongside CorpusIQ, a third-party server like this one can be combined with a company's own business data in a single answer.