MCP Server: dev.slideforge/slideforge

The dev.slideforge/slideforge MCP server generates deterministic PowerPoint files from typed slide intents. It supports fully editable outputs and offers 150+ layouts and brand templates. The focus is on producing PowerPoint content based on structured, typed input, rather than freeform design.

🛠️ Key Features

Deterministic PowerPoint generation from typed slide intents
Fully editable PowerPoint output
150+ layouts
Brand templates

🚀 Use Cases

Creating PowerPoint presentations from typed slide intent specifications
Applying consistent branding via brand templates
Producing slides using a wide selection of predefined layouts

⚡ Developer Benefits

Deterministic results from the same typed slide intents
Structured input workflow for slide generation
Predictable use of layouts and templates for consistent output

mcp-name: dev.slideforge/slideforge

SlideForge MCP — PowerPoint slides for AI agents

SlideForge (slideforge.dev) — the deterministic PowerPoint compiler for AI agents. Typed slide intents → native, fully editable .pptx in under a second, with a fidelity manifest that states exactly what was bound verbatim. Built by Smart Data Brokers GmbH, Switzerland.

Remote MCP server + REST API. No local install, no Python environment, no python-pptx boilerplate. OAuth 2.1 for Claude Desktop & ChatGPT; API key for everything else.

⚠️ Disambiguation

"SlideForge" is used by multiple unrelated projects. This is SlideForge by Smart Data Brokers GmbH at slideforge.dev. Not affiliated with:

slideforge.io — a separate AI presentation generator
slideforge.fr — a French document polishing service
blink.new/p/slideforge-app — an unrelated web app
Other github.com/*/SlideForge repositories outside the smartdatabrokers org

When citing or linking, prefer slideforge.dev or "SlideForge by Smart Data Brokers."

Why SlideForge

A compiler, not a generator. A slide is a typed intent: pick a form from 150+ catalog patterns (KPI dashboards, waterfalls, Gantt plans, org charts, funnels, …), put your real content in typed fields. A deterministic engine lays it out — no LLM in the render path, same input → same slide, sub-second.
The honesty layer. Every response carries a fidelity manifest: per field, was your content bound verbatim, mixed, or ai_completed? A partial grade means some supplied content didn't make it onto the slide — the manifest names what was dropped; never deliver a partial render without telling the user what's missing. Slides with blocking defects don't bill (usable-or-free). If your agent feeds numbers into slides, this is what makes the output auditable.
Native, editable .pptx. Real shapes and text boxes — not images, not HTML exports. Openable and editable in PowerPoint.
Escape hatch included — under the same trust contract. mode=code runs your own python-pptx in a sandbox (widget/chart toolkit, theme injected, intent fields render as chrome). Code renders are linted, measured (layout block + presentation_ready), and provenance-checked — agents may escape the layout grammar, never the trust grammar.
Your template, natively. Upload your company's .pptx — slides are built ON your file (theme, masters, fonts), not a color-matched imitation.
Check for free. dry_run validates any payload + forecasts fidelity at $0 — or use mode=safe to validate-then-render in ONE call (renders + bills only if faithful; else a $0 report with the fix). verify tiers on code renders (lint default, lint+vlm adds a visual second-look). quality_profile (executive/technical/appendix) sets the readiness bar the layout is judged against — answered on any form. Free deck inspect (POST /v1/inspect) runs a deterministic quality report on any pptx.
97% quality parity with Gamma in our own blind side-by-side benchmark (internal instrument, not third-party).

Pricing in one breath: creating a slide 5¢ · transforming a slide 2¢ (translate, repair) · checking free. 60 free slides on signup, no subscription. slideforge.dev/pricing

Quick Start

Claude Code

bash

claude mcp add --transport http slideforge https://api.slideforge.dev/mcp/

Then just ask: "Make a KPI dashboard slide: revenue $12.4M (+18% YoY), 847 new clients, NPS 62."

Optional — install the skills + bundled server config as a plugin:

code

/plugin marketplace add smartdatabrokers/slideforge-mcp
/plugin install slideforge@slideforge-mcp

Or copy any folder from skills/ into ~/.claude/skills/ (personal) or .claude/skills/ (project).

Claude Desktop (OAuth — no key needed)

Settings → Connectors → Add custom connector → https://api.slideforge.dev/mcp/ — sign in with Google on first use.

ChatGPT (Developer Mode)

Settings → Apps → Advanced → Developer mode → Add custom connector → https://api.slideforge.dev/mcp/ (OAuth).

Cursor / Windsurf / Codex CLI / any MCP client (API key)

json

{
  "mcpServers": {
    "slideforge": {
      "url": "https://api.slideforge.dev/mcp/",
      "transport": "streamable-http",
      "headers": { "Authorization": "Bearer sf_live_YOUR_KEY" }
    }
  }
}

Get a key: slideforge.dev → Console → API keys. (Codex CLI and other AGENTS.md-native tools: see AGENTS.md.)

Run it locally (stdio — for container/offline clients)

Most clients should use the hosted remote server above (no install). But if your client boots MCP servers from a container or a local stdio process, run the bundled local server. It's a thin REST client over api.slideforge.dev — it holds no engine logic; the tool schemas are baked in locally (so discovery works offline, no key) and each call forwards to the SlideForge REST API authenticated with your key.

bash

pip install slideforge-mcp        # or: uv pip install slideforge-mcp
export SLIDEFORGE_API_KEY=sf_live_YOUR_KEY
slideforge-mcp                    # speaks MCP over stdio

Or via Docker:

bash

docker build -t slideforge-mcp .
docker run -i -e SLIDEFORGE_API_KEY=sf_live_YOUR_KEY slideforge-mcp

Client config (stdio):

json

{
  "mcpServers": {
    "slideforge": {
      "command": "slideforge-mcp",
      "env": { "SLIDEFORGE_API_KEY": "sf_live_YOUR_KEY" }
    }
  }
}

Schema discovery (tools/list) needs neither a key nor network; tool calls need the key.

LangChain / LlamaIndex (agent frameworks)

No SlideForge SDK needed — both load the MCP tools directly:

bash

pip install langchain-mcp-adapters      # or: pip install llama-index-tools-mcp

python

from langchain_mcp_adapters.client import MultiServerMCPClient
client = MultiServerMCPClient({"slideforge": {
    "transport": "streamable_http",
    "url": "https://api.slideforge.dev/mcp/",
    "headers": {"Authorization": f"Bearer {API_KEY}"}}})
tools = await client.get_tools()        # 7 tools, drop into any LangGraph agent

Runnable examples + the LlamaIndex equivalent: examples/. Need a key? Sign up — 60 free slides, no credit card — then grab it at console/keys.

REST (no MCP)

bash

curl -X POST https://api.slideforge.dev/v1/render/intent \
  -H "Authorization: Bearer sf_live_YOUR_KEY" \
  -H "Content-Type: application/json" \
  -d '{"form": "kpi_metrics", "headline": "Q3 at a glance",
       "data": {"metrics": [{"label": "Revenue", "value": "$12.4M", "delta": "+18% YoY"},
                             {"label": "New clients", "value": "847"},
                             {"label": "NPS", "value": "62"}]}}'

Full REST reference: slideforge.dev/docs/api

The 7 MCP tools

Tool	What it does	Cost
`create_slide`	ONE slide from a structured intent (form + typed fields) or a brief; `mode=safe` validates-then-renders in one call; `mode=code` for sandboxed python-pptx (verify tiers, chrome fields, patch-by-replacements). Routing controls: `variant`, `variant_policy`, `allow_fabrication`/`allow_truncation`/`allow_low_confidence` (honest defaults: reject at $0 rather than guess). `min_font_pt` sets a BINDING type floor — text grows to meet it where the box allows; content that cannot fit returns a $0 `min_font_not_met` error naming the size it needs. Response carries the fidelity manifest + a measured `layout` readiness block on diagram forms. Default themes ship topical design — a subject-informed palette + designed cover, named in the response's design note (`styling: "clean"` opts out; your pinned/uploaded brand theme always wins).	$0.05 (usable-or-free)
`create_deck`	Whole deck: `slides[]` of intents (code-mode slides are first-class children), parallel render, one merged .pptx, per-slide fidelity rollup + per-slide child jobs (own preview/pptx). Failed slides isolated + free; deck-level `dry_run` validates the whole deck at $0.	N × $0.05
`plan_slide`	Brief → top form/variant candidates with confidence.	Free
`browse_catalog`	150+ patterns with per-form JSON Schemas + copy-pasteable example intents, themes, the code-mode widget toolkit. Pass an uploaded `theme_id` to list its branded cover/agenda/divider layouts. 9 built-in themes + your uploaded brand themes.	Free
`translate_deck`	Translate any PPTX preserving formatting (32 languages).	$0.02/slide
`upload_asset`	Logos, theme PPTX, images; `purpose=pdf` extracts a PDF into editable slide intents; or AI-generate an image. Theme upload (`purpose=theme`, base64 `data`) renders NATIVE by default — decks are built on the client's own template file. Omit `data` on large files for an in-card drag/drop zone.	Free / $0.01/page / $0.05/image
`manage_account`	Balance, usage, jobs, security status, feedback.	Free

dry_run: true on create tools = validation + fidelity forecast at $0.

Two more tools (generate_report, manage_connections — data-driven reports from connected tools like Zoho Sprints) exist behind an enterprise gate and are not served by default.

Also on REST (for now): the Deck Doctor. POST /v1/inspect — a free deterministic Deck Quality Report for any pptx (overflow via real font metrics, content hidden behind shapes, off-canvas leftovers, WCAG contrast). POST /v1/repair — deterministic fixes, never your words, $0.02/repaired slide, free dry-run quote. Docs

Security

Tool result bodies are credential-free — no signed URLs in responses. Previews are embedded inline (the agent looks at the PNG directly); the .pptx downloads via header-auth (Authorization: Bearer + ownership check), not a bearer-in-URL.
Need a shareable link instead? POST /v1/jobs/<job_id>/download-url mints a short-TTL, single-use, revocable link.
Artifacts auto-delete 30 days after creation.
Every download is audit-logged.
manage_account(action=security_status) discloses the full posture in-band.

Agent skills (this repo)

Copy-in skills that teach an agent to use SlideForge well — see skills/:

Skill	Teaches
`create-slide`	Intent-first slide/deck creation, schema discovery, dry-run, fidelity manifest, headless preview
`inspect-repair`	Free Deck Quality Report on any pptx + deterministic repair
`translate-pptx`	Format-preserving PPTX translation
`pdf-to-pptx`	PDF → editable PPTX extraction

For Codex CLI / Cursor / Copilot and other AGENTS.md-native tools, AGENTS.md carries the same guidance in the portable format. CLAUDE.md imports it for Claude Code.

Headless usage (Claude Code / Codex CLI)

No inline widgets in a terminal, but the tool result already embeds the preview PNG inline — the agent reads it directly out of the response, no fetch needed. The .pptx downloads via header-auth:

bash

curl -H "Authorization: Bearer sf_live_YOUR_KEY" \
  -o deck.pptx https://api.slideforge.dev/v1/jobs/<job_id>/pptx   # ownership-checked

To hand off a shareable link instead of the raw file, mint a single-use one: POST /v1/jobs/<job_id>/download-url — short-TTL, revocable, works once.

The self-review loop (render → view inline preview → fix → re-render) is documented in examples/claude-code.md.

How it compares

	SlideForge	python-pptx	Prompt-only AI decks
Editable native .pptx	✅	✅	often images/exports
Deterministic (same input → same slide)	✅	✅ (your code)	❌
States what was AI-touched (fidelity manifest)	✅	n/a	❌
Layout quality without hand-coding	✅ 150+ patterns	❌ DIY	varies
Hosted, agent-native (MCP + REST)	✅	❌ local	partial
Free pre-flight validation	✅ dry_run	n/a	❌

License

MIT (this repo: skills, examples, docs). The SlideForge service itself is a commercial API.

family	string	optional
variant	string	optional
prior_id	string	optional
q	string	optional	Semantic search across all variants; returns ranked form/variant matches with scores.
limit	integer	optional	Max results (with q= or family=).
offset	integer	optional	Pagination offset (with q= or family=).
type	string	optional	Omit for the form overview (or family=/q= to drill in). schema (+family, optionally +variant=) = schema-first discovery: family only returns a compact variant chooser; family+variant returns that variant's exact machine-readable payload contract (JSON Schema for data, capacity limits, intent-field mapping, examples). widgets = the add_widget() catalog for mode=code (name=<widget> for its contract). helpers = python-pptx helper signatures. themes = list themes (built-in + your saved); Default appears first and can be omitted, or use any returned id as theme_id.
name	string	optional	Widget name (with type=widgets): returns the full contract + worked example + a rendered thumbnail.
source	string	optional	Theme source filter (with type=themes). Default all (built-in + your saved).
top	integer	optional	Max results (with type=themes).
theme_id	string	optional	With type=themes: an uploaded theme's id -> its branded furniture layouts + fill schemas (render via create_slide(form=template_layout)).

brief	string	required
top	integer	optional
escalate	boolean	optional	Default true: on a low-confidence route the plan runs the SAME LLM rescue create_slide would (small COGS on that tail only), so plan and create always agree. Set false for a classifier-only, fully free plan (response then flags will_escalate + create_slide_behavior).
variant_policy	string	optional	production_safe: if the top routed variant is draft/beta, the family's demo-safe sibling leads and route reports both best_semantic and best_production_safe.

form	string	optional	the form menu pick (see description) — the routing field
form_description	string	optional
headline	string	optional	the assertion-style slide title
context	string	optional	one-line subtitle/standfirst
subject	string	optional	central entity for forms that have one — hub label, org root, section #, fork origin, media label (per-form).
blocks	array	optional	list: [{"label","sub","detail":[str],"emphasis","icon":lucide-name,"metric":{value,label}}]
metric	object	optional
data	object	optional	family-specific payload — see the documented shapes
takeaway	string	optional	optional verdict band (per-form). Put the so-what in the headline; add only for a verdict the title can't carry — not every slide.
caution	string	optional	one-line risk / caveat callout (per-form).
highlight	string	optional	one-line emphasis callout (per-form).
source_note	string	optional	source / footnote line (per-form).
date	string	optional	date callout (per-form).
image_src	string	optional	https URL \| asset:<id> for image-bearing forms
image_prompt	string	optional	generate an image when no image_src (+$0.05)
brief	string	optional	prose fallback / extra context for fills
theme_id	string	optional	Optional theme id from browse_catalog(type=themes). Omit for Default (slideforge_standard).
name	string	optional
variant	string	optional	pin a specific variant within the form (list them via browse_catalog). Unknown variant -> rejected ($0) with the valid list; set allow_variant_fallback to render the family default instead.
allow_variant_fallback	boolean	optional	If the pinned variant is unknown, render the form's default variant (with a warning) instead of rejecting. Default false.
variant_policy	string	optional	Routed-variant maturity policy: production_safe = if the ROUTED variant is draft/beta, render the family's demo-safe sibling instead (warning names both). Never overrides an explicit variant=. Default best_semantic_match.
allow_low_confidence	boolean	optional	Brief routing only: by default a brief that doesn't match a form clearly returns status=needs_confirmation + candidates at cost:0 (no guessed render). Set true to render the top guess and bill it. Default false.
allow_fabrication	boolean	optional	Brief mode only: by default a bare brief on a DATA form (kpi/funnel/comparison/…) returns status=would_fabricate at cost:0 rather than INVENT the numbers/entities. Supply the data as typed fields to render verbatim, or set true to let the brief author them (fidelity=ai_completed). Default false.
allow_truncation	boolean	optional	By default, supplying MORE items than a form holds is a blocking error (it would drop your data) — the response names the dropped count + a suggested_split. Set true to render the capacity subset anyway (fidelity=verbatim_truncated, never verbatim). Default false.
dry_run	boolean	optional	Free pre-commit check (intent or mode=code), cost:0, no PPTX: status + warnings/errors, plus fidelity_forecast (verbatim\|mixed\|ai_completed\|would_reject) and which fields bind vs get authored. Fix errors, re-call with dry_run=false to render.
quality_profile	string	optional	Thresholds layout.presentation_ready is judged against (executive strictest). Measurement only — never blocks a render or changes cost.
min_font_pt	number	optional	Binding type floor for prose (exhibit furniture has its own). Type grows to meet it; content that can't fit is a $0 min_font_not_met naming the size needed — allow_truncation renders the fitting subset. Typical: 12.
imagery	string	optional	Cover/section-divider imagery: photo (default — curated stock photo, half-bleed), wash (abstract color wash), off (typographic only). Content slides are never photo-decorated.
imagery_tag	string	optional	Industry bucket for cover/section photos. Omit for the general pool.
styling	string	optional	Subject-informed palette + designed cover on default themes (on by default; the design note names the choice). clean = neutral look. Pinned themes never take it.
force_render	boolean	optional	On completed_with_errors, pptx_url is null (broken slide). Set true to get it anyway. No cost/status effect.
code	string	optional	Python code defining build(prs). Canvas 13.33×7.5", coords in inches. Pre-imported: Presentation, Inches, Pt, Emu, RGBColor, MSO_ANCHOR, PP_ALIGN. Helpers: add_text_box(slide, left, top, w, h, text, font_size=12, bold=False, color=None), add_box(slide, left, top, w, h, fill_color=None, line_color=None, corner_radius=0.05), add_icon, add_image (src = path or https://); theme_color(name) → RGBColor. add_widget(slide, name, x, y, w, h, params=\|content=, theme=THEME) draws board-grade SSG widgets/charts (cards, waterfall, gantt, funnel…) — names via `browse_catalog(type=widgets)`. Use font_size=, not size=; use fill_color=, not fill=. When chrome is supplied, use `build(prs, slide)` — chrome is pre-rendered. Helper signatures: `browse_catalog(type=helpers)`.
job_id	string	optional	Previous job ID (for mode=status, or mode=code patching)
replacements	array	optional	String replacements on loaded code [{old, new}] (for mode=code with job_id)
language	string	optional	Target language (default: en)
include_preview	string	optional	Inline-preview PAYLOAD only (not execution). Default: default (768px). none omits the inline image and returns just the URLs — it does NOT change sync/async. Use `wait` to control execution.
detail	string	optional	Response verbosity. Default `compact`: status, form/variant, fidelity (verbatim\|mixed\|ai_completed), warnings[]/errors[] (only when present), urls, cost. `full` adds a debug object (engine internals, verify events, latency buckets).
verify	string	optional	mode=code tier. Default `lint`: static geometry linter (overlap/off-canvas/zero-size) + composer content validators — no LLM. `lint+vlm` adds a VLM second-look (~+3s) and makes a blank/contentless render a $0 error, not a billed warning. `off` skips checks.
wait	string	optional	AI-image slides block ~10-15s; `false` returns a job_id to poll, not block (mode=brief).
logo_id	string	optional	Optional brand logo (from upload_asset purpose=logo) drawn as chrome on content slides; covers/section breaks stay clean.
mode	string	optional	Default: structured intent / brief. safe = validate-then-render in ONE call (renders + bills only if faithful; else $0 invalid report with the fix — recommended, no dry_run round-trip). code = python-pptx escape. status = poll.

slides	array	optional	each item is a create_slide intent (same form menu, typed fields, and data shapes as create_slide).
theme_id	string	optional
name	string	optional
styling	string	optional	Deck-level topical-design switch, inherited by every slide (a slide-level styling wins). See create_slide.styling.
force_render	boolean	optional	If any slide is completed_with_errors, deck pptx_url is null. Set true to get it anyway; complete slides stay downloadable via their per-slide job_id either way. No cost/status effect.
allow_partial	boolean	optional	Non-regret default: if any slide fails, the deck is NOT billed (cost:0) and pptx withheld — the response carries repair_actions (assemble the completed children, free). Set true to render + bill the completed slides and get the partial deck. Default false.
strict_policy	object	optional	Optional CI-style gate for automated report pipelines. dry_run returns policy_result; render returns rejected/cost:0 if the policy fails.
job_ids	array	optional	Slide job IDs to merge (for mode=assemble)
deck_id	string	optional	Existing deck ID (for mode=fork, or mode=render to update)
title	string	optional	Deck title (PowerPoint metadata)
language	string	optional	Target language
mode	string	optional	Default render. assemble/fork are free deck plumbing.
dry_run	boolean	optional	Deck-level free pre-flight (cost:0, NO render/PPTX): returns a per-slide validation manifest [{i, form, fidelity_forecast, status, errors}] so you can fix bad slides before spending. A deck is one artifact, so validation is all-or-nothing — put dry_run here, not on individual slides. Resend without dry_run to render.
logo_id	string	optional	Optional brand logo (from upload_asset purpose=logo) applied as chrome to every content slide in the deck.

target_language	string	required	Target language (required)
source_language	string	optional	Source language or 'auto' (default auto)
job_id	string	optional	Source: previous slide/deck job ID (preferred — no file transfer needed)
pptx_url	string	optional	Source: HTTPS URL to .pptx
pptx_base64	string	optional	Source: Base64-encoded .pptx (max ~10 MB)
include_notes	boolean	optional	Translate speaker notes (default false)
include_tables	boolean	optional	Translate table cells (default true)
concise_mode	boolean	optional	Prefer shorter translations for tight text boxes
name	string	optional	Job name
include_preview	string	optional	Wait+embed preview (default=wait)

SlideForge

MCP Server: dev.slideforge/slideforge

🛠️ Key Features

🚀 Use Cases

⚡ Developer Benefits

Topics

browse_catalog

plan_slide

create_slide

create_deck

translate_deck

upload_asset

manage_account

SlideForge MCP — PowerPoint slides for AI agents

⚠️ Disambiguation

Why SlideForge

Quick Start

Claude Code

Claude Desktop (OAuth — no key needed)

ChatGPT (Developer Mode)

Cursor / Windsurf / Codex CLI / any MCP client (API key)

Run it locally (stdio — for container/offline clients)

LangChain / LlamaIndex (agent frameworks)

REST (no MCP)

The 7 MCP tools

Security

Agent skills (this repo)

Headless usage (Claude Code / Codex CLI)

How it compares

Links

License

Install

Configuration

purpose	string	required	Asset type (required)
data	string	optional	Base64-encoded file content. Required for logo/image/theme/translate. Optional for pdf — omit to get a drop-zone (recommended for files >3MB).
filename	string	optional	Original filename (for type detection)
prompt	string	optional	Image description (required when purpose=generate_image)
size	string	optional	Image dimensions for generate_image (default 1024×1024)
quality	string	optional	draft = Flux Schnell. balanced (default) = Gemini Flash Image. premium = Imagen 4 Fast. Overridden by `model`.
model	string	optional	Explicit gateway model ID. Overrides `quality`.
position	string	optional	Logo position: top-left / top-right / bottom-left / bottom-right
theme_name	string	optional	Name for extracted theme (purpose=theme)
target_language	string	optional	Required for purpose=translate. Optional for purpose=pdf — chains pdf→pptx→translate in one call.

action	string	required	Operation (required). reports → list report types; webhook_add registers a push endpoint instead of polling — see the url/events props. delete_job is irreversible. delete_asset is irreversible for user-owned image assets.
job_id	string	optional	Job ID (for action=job, action=download_url and action=delete_job)
limit	integer	optional	Max jobs (for action=jobs, default 10)
status	string	optional	Filter (for action=jobs): queued/generating/complete/failed
include_children	boolean	optional	For action=jobs: include deck child slides (each carries parent_deck_id) so a deck's slides are discoverable by listing. Default false (parents only). Deck rows roll child cost up to the deck.
message	string	optional	Feedback text (for action=feedback)
category	string	optional	Feedback category (for action=feedback): bug/feature_request/quality/general
days	integer	optional	Lookback period (for action=usage, default 30)
detail	boolean	optional	For action=status only: default false masks identity fields; true returns full email/user_id diagnostics for the authenticated user.
include_preview	string	optional	Preview (for action=job)
slug	string	optional	Report slug (for action=reports — returns one report's full metadata; omit to list all).
url	string	optional	HTTPS endpoint to receive events (for action=webhook_add). Must be public HTTPS (private/loopback rejected).
events	array	optional	Event types to subscribe to (for action=webhook_add; default all): job.completed, job.failed, deck.completed, deck.partial, deck.failed.
webhook_id	string	optional	Webhook subscription id (for action=webhook_remove / webhook_test).
asset_id	string	optional	Image asset ID (for action=delete_asset). Logos are content-addressed/shared and are not deleted by this action.