tesserae-mcp
Server Configuration
Describes the environment variables required to run the server.
| Name | Required | Description | Default |
|---|---|---|---|
| TESSERAE_URL | No | The URL of the Tesserae instance (default: http://127.0.0.1:8765) | http://127.0.0.1:8765 |
| TESSERAE_MCP_TOKEN | No | The MCP API token from Tesserae settings. Omit if running on the same machine (loopback trusted). |
Capabilities
Features and capabilities supported by this server
| Capability | Details |
|---|---|
| tools | {
"listChanged": false
} |
| prompts | {
"listChanged": false
} |
| resources | {
"subscribe": false,
"listChanged": false
} |
| experimental | {} |
Tools
Functions exposed to the LLM to take actions
| Name | Description |
|---|---|
| list_widgetsA | List every widget that can be placed on a canvas (with its fragments) and the available theme/style/font appearance options. |
| get_widget_optionsA | Get the configurable options for one widget, so you can fill an element's "options" correctly (e.g. a weather widget's location). Each option carries a "format" hint for its type. Big choice lists (HA entity pickers) are omitted by default (the option shows "choices_count" + a "choices_endpoint"); pass include_choices=True to inline them, or call get_widget_choices() to page. |
| get_widget_choicesA | Page through the choice rows for one of a widget's options (kept out of get_widget_options so a picker with hundreds of entries doesn't bloat the schema). "q" filters by case-insensitive substring on value/label. |
| probe_widget_dataA | Return a widget's data as JSON to pick "field" paths before binding a data primitive. Returns {data, data_source, reason, fields}: "data_source" is "live" (real fetch), "sample" (demo fallback because nothing was configured), or "error" (fetch failed) so you never mistake a placeholder for a real result; "fields" lists the bindable dot-paths with sample values (a wrong key simply isn't in the list; an empty payload has fields with null values). |
| list_devicesA | List registered display devices with panel dims AND colour capability: "color_mode" (e.g. "6-colour (Spectra 6)"), "colors" (the renderable palette as hex), "gamut", "orientation", and a "mono" flag. Match a canvas's w/h to the target panel, and design within its palette so colours don't quantise away on the hardware. |
| list_pagesA | List existing canvas (freeform) dashboards. |
| create_canvas_pageA | Create a new, empty canvas dashboard and return its id. Then set_canvas() to lay it out. Size it to your target panel (see list_devices). |
| get_canvasA | Get the full canvas document (size, appearance, and every element) for a page, plus "rev" / "updated_at" / "updated_by". Keep the "rev" and pass it as base_rev on your next write to be warned (HTTP 409) if the page drifted. |
| update_elementA | Change ONE element in place without re-sending the whole document. "patch" is a partial element ({field: value, ...}) merged over the existing one (a provided "options"/"parts" replaces wholesale). The cheap edit path for a big canvas: change a precision, a colour, a location, one box. Returns the ack. |
| delete_elementB | Remove ONE element from a canvas by id. Returns the compact ack. |
| patch_canvasA | Change document-level fields (any of name, w, h, theme, style, font, bg, bg_image, bg_fit) without touching the elements. Use update_element / set_canvas for elements. Returns the ack. |
| arrangeA | Compute "count" aligned child boxes inside "box" ({x,y,w,h}) for a "grid" / "row" / "column" layout, so you place cells by intent instead of hand-computing pixels. "gap" is the space between cells, "pad" the inset from the box edge, "cols" forces a grid column count (default ~sqrt). Returns {boxes:[{x,y,w,h}, ...]}; spread them across your elements' geometry (bake them in as normal elements — they stay individually editable). |
| measure_textA | Measure how wide/tall text renders in a widget font, so a box fits its content (prevents clipping). "items" is a list of {text, font?, size?, weight?, max_width?}. Returns {items:[{text,width,height,fits}]} where "fits" is whether the text is within max_width. Font names come from list_widgets().appearance. |
| render_reportA | Read back what a canvas actually rendered, as JSON (a companion to render_preview's image). Per element: the resolved box, the text that rendered, overflow/clip flags (overflow_x when content is wider than its box), "data_source" (live | sample | error | static), and computed colours; plus the board's resolved background / theme. Use it to verify a render — catch clipping, confirm live data, read the real colours — without parsing a PNG. (Widget cells render into shadow DOM, so their "text" may be empty; data primitives and decorations report their text.) |
| render_previewA | Render the canvas to a PNG at its authored size and return the image, so you can visually check the layout and iterate. This is your feedback loop: place → render_preview → adjust → set_canvas → render_preview again. For a machine-readable check (values, overflow, colours), use render_report(). |
| push_to_deviceA | Render the canvas and push it to the given devices (ids from list_devices). device_ids is required — pushing is always explicit. |
| set_canvasA | Replace a canvas dashboard's document. Returns a compact {ok,id,rev,elements} ack (not the full document), or an error with field-level "details" (HTTP 422) if the document is invalid, so you can correct it and retry. Pass base_rev (the rev from get_canvas) to be warned with HTTP 409 if the page changed under you. For a one-field change prefer update_element / patch_canvas. After setting, call render_preview() (or render_report()) to check the result. A canvas document is JSON: { "w": int, "h": int, # artboard size in px (match the target panel) "theme": str, "style": str, # appearance ids from list_widgets().appearance "font": str, "bg": str, # optional font id and background colour override "els": [ , ... ] # painted in list order: first = back, last = front } Elements may sit partly off the panel (it clips at the edge). Each element has a unique "id" and a box "x","y","w","h" (px, top-left origin; x/y may be negative), plus optional "opacity" (0-100) and "rotate" (degrees). By "kind":
LIVE BINDINGS ("bind" on ANY element -- makes a SHAPE reflect data): Data elements auto-update, but shapes (rect/ellipse/icon/line/text) are static geometry. Add "bind": [ , ... ] to drive a shape's props from data each render (in lockstep with data elements, no polling). A binding is: {"source":"","options":{...},"field":"","transform":"","params":{...}} transforms: position -- scalar to a coordinate: {"axis":"x"|"y","in":[lo,hi],"out":[p0,p1],"center":} lo/hi may be numbers OR other field paths (e.g. "sun.riseMin"). length -- scalar to a size: {"dim":"w"|"h","in":[lo,hi],"out":[minPx,maxPx],"anchorMax":?} pick -- integer field indexes arrays: {"set":{"x":[...],"color":[...],...},"center":?} color -- scalar to a colour by ascending thresholds: {"stops":[[max,"#hex"],...],"else":"#hex"} gradient -- scalar interpolated smoothly along colour stops: {"stops":[[value,"#hex"],...]} (quantised to the panel palette on e-ink; a value-driven gradient, not animation) icon -- code/string to a Phosphor glyph: {"table":{"":"ph-name"},"default":"ph-name"} Several bindings combine (e.g. bind x by position AND colour by threshold). A binding that can't resolve its value is skipped, so the element keeps its authored props. FIELD PATHS ("field" on data elements): dotted -- "current.temp" array index -- "hourly.0.temp" (or "hourly[0].temp") pluck (for charts) -- "series.*.total" (or "series[].total") maps .total over every item of the array "series", yielding an array of numbers. Charts (line/bar/sparkline) need a field that resolves to an array of numbers -- use pluck to get one from an array-of-objects. probe_widget_data() returns a "fields" list of the bindable paths (with sample values), so you don't have to reverse-engineer the shape. EDITING WITHOUT RESENDING EVERYTHING: update_element / delete_element / patch_canvas change one thing without re-sending the whole document (cheaper, fewer errors on a big canvas). add_element appends one. AVOID CLOBBERING A CONCURRENT EDIT: get_canvas() returns a "rev". Pass it as base_rev to a write; if the page changed since (someone edited it in the UI, or another agent), the write returns HTTP 409 with the current rev, so you re-read instead of overwriting. LAY OUT BY INTENT, NOT PIXELS: arrange(box, layout, count) returns aligned child boxes (grid/row/column) to spread across your elements -- no hand-computed x/y. measure_text() tells you how wide text renders so a box fits its content (prevents clipping). render_report() reads back what actually rendered (per-element value, overflow flags, live-vs-sample, colours) so you verify without eyeballing the PNG. MATCH THE HARDWARE: list_devices() reports each panel's colour capability (color_mode, the renderable palette as hex, and a "mono" flag). Design within that palette so colours don't quantise away on the panel. |
| add_elementA | Append ONE element to a canvas and save (each call is a separate save, so an open editor updates live as you build). 'element' is a single element object; returns {ok,id,rev,elements,element_id}. Use set_canvas to replace the whole layout at once. A canvas document is JSON: { "w": int, "h": int, # artboard size in px (match the target panel) "theme": str, "style": str, # appearance ids from list_widgets().appearance "font": str, "bg": str, # optional font id and background colour override "els": [ , ... ] # painted in list order: first = back, last = front } Elements may sit partly off the panel (it clips at the edge). Each element has a unique "id" and a box "x","y","w","h" (px, top-left origin; x/y may be negative), plus optional "opacity" (0-100) and "rotate" (degrees). By "kind":
LIVE BINDINGS ("bind" on ANY element -- makes a SHAPE reflect data): Data elements auto-update, but shapes (rect/ellipse/icon/line/text) are static geometry. Add "bind": [ , ... ] to drive a shape's props from data each render (in lockstep with data elements, no polling). A binding is: {"source":"","options":{...},"field":"","transform":"","params":{...}} transforms: position -- scalar to a coordinate: {"axis":"x"|"y","in":[lo,hi],"out":[p0,p1],"center":} lo/hi may be numbers OR other field paths (e.g. "sun.riseMin"). length -- scalar to a size: {"dim":"w"|"h","in":[lo,hi],"out":[minPx,maxPx],"anchorMax":?} pick -- integer field indexes arrays: {"set":{"x":[...],"color":[...],...},"center":?} color -- scalar to a colour by ascending thresholds: {"stops":[[max,"#hex"],...],"else":"#hex"} gradient -- scalar interpolated smoothly along colour stops: {"stops":[[value,"#hex"],...]} (quantised to the panel palette on e-ink; a value-driven gradient, not animation) icon -- code/string to a Phosphor glyph: {"table":{"":"ph-name"},"default":"ph-name"} Several bindings combine (e.g. bind x by position AND colour by threshold). A binding that can't resolve its value is skipped, so the element keeps its authored props. FIELD PATHS ("field" on data elements): dotted -- "current.temp" array index -- "hourly.0.temp" (or "hourly[0].temp") pluck (for charts) -- "series.*.total" (or "series[].total") maps .total over every item of the array "series", yielding an array of numbers. Charts (line/bar/sparkline) need a field that resolves to an array of numbers -- use pluck to get one from an array-of-objects. probe_widget_data() returns a "fields" list of the bindable paths (with sample values), so you don't have to reverse-engineer the shape. EDITING WITHOUT RESENDING EVERYTHING: update_element / delete_element / patch_canvas change one thing without re-sending the whole document (cheaper, fewer errors on a big canvas). add_element appends one. AVOID CLOBBERING A CONCURRENT EDIT: get_canvas() returns a "rev". Pass it as base_rev to a write; if the page changed since (someone edited it in the UI, or another agent), the write returns HTTP 409 with the current rev, so you re-read instead of overwriting. LAY OUT BY INTENT, NOT PIXELS: arrange(box, layout, count) returns aligned child boxes (grid/row/column) to spread across your elements -- no hand-computed x/y. measure_text() tells you how wide text renders so a box fits its content (prevents clipping). render_report() reads back what actually rendered (per-element value, overflow flags, live-vs-sample, colours) so you verify without eyeballing the PNG. MATCH THE HARDWARE: list_devices() reports each panel's colour capability (color_mode, the renderable palette as hex, and a "mono" flag). Design within that palette so colours don't quantise away on the panel. |
Prompts
Interactive templates invoked by user choice
| Name | Description |
|---|---|
No prompts | |
Resources
Contextual data attached and managed by the client
| Name | Description |
|---|---|
No resources | |
Latest Blog Posts
- Who's Calling? MCP Hosts Are an Identity Blind Spot (And the Spec Knows It)By Om-Shree-0709 on .mcpAgent IdentityOAuth 2.1
- Your AI Chatbot Just Exposed Your CEO's Salary to an InternBy Om-Shree-0709 on .Agent IdentityMCP SecurityOAuth Delegation
- Why MCP Servers Need Execution Sandboxing (And Why Your Current Stack Isn't Enough)By Om-Shree-0709 on .Agentic AiPrompt InjectionWebAssembly
MCP directory API
We provide all the information about MCP servers via our MCP API.
curl -X GET 'https://glama.ai/api/mcp/v1/servers/dmellok/tesserae-mcp'
If you have feedback or need assistance with the MCP directory API, please join our Discord server