Everything your agent can do on one key.

17 tools over one MCP endpoint — images, speech, web search, memory and publishing.

Show

Image

3· Draw it, restyle it, or read what is already there.

Image generationLive

image.generate

Turn a prompt into a picture.

Render artwork, product shots, diagrams, and icons straight from a text prompt. Pick an aspect ratio and output format, or ask for several variations at once.

Image editingLive

image.edit

Restyle or extend images you already have.

Pass one or more reference images and a prompt to restyle, extend, or recombine them. Available over MCP only — the REST image API generates from a prompt alone.

Image understandingLive

image.describe

Eyes for your agent — read any screenshot, chart, or photo.

Get a detailed description of an image: scenes, objects, charts, UI screenshots. Pass your own prompt to steer what the model looks for.

Audio

2· A voice to speak with and ears to listen.

Text to speechLive

audio.synthesize

Give your agent a voice, in the voice you pick.

Turn any text into natural spoken audio with a selectable voice, format, and speed. Built for voice agents, narration, accessibility, and notifications.

Speech to textLive

audio.transcribe

Ears for your agent — meetings, voice notes, calls.

Transcribe recordings into clean text. Upload a file directly or point at a URL, and optionally hint the spoken language.

Web

1· Live results, past the training cutoff.

Web searchLive

search.web

Real-time results, so answers aren't stuck at the training cutoff.

Run a live web search and get back ranked results the agent can read and cite. Cap the result count per call.

Memory

5· What the agent learned, still there next session.

Save a memoryLive

memory.save

Write a markdown note under a key the agent chooses.

Create or overwrite one entry. Each key belongs to the calling key by default, or to a shared account-wide space when the key has memory sharing on.

Recall a memoryLive

memory.get

Fetch one entry by its exact key.

Read back a single stored note verbatim. Costs nothing — memory reads are unmetered.

List memoriesLive

memory.list

Every key in scope, with a preview of each.

Enumerate what the agent has stored so it can decide what to read in full. Unmetered.

Search memoryLive

memory.search

Ask in plain language; an LLM ranks what's stored.

Natural-language retrieval over everything the agent saved. There is no embedding index — a model reads the candidates and ranks them, so a query works from the first note onward.

Forget a memoryLive

memory.delete

Drop one entry by key.

Remove a single note. Unmetered, and immediate.

Artifacts

6· Publish the work as a link you can send.

Publish a siteLive

artifact.deploy

Ship a zipped static site and get a shareable URL back.

Upload a zip with an index.html at its root and Router9 hosts it on its own origin, with SPA fallback. Public, or private behind a one-time share link.

List artifactsLive

artifact.list

Every site you've deployed, with URL and visibility.

Enumerate the account's live artifacts, plus usage against the plan's storage and count limits. Unmetered.

Get an artifactLive

artifact.get

One deployed site by key.

Read a single artifact's URL, name, visibility, size, and view count. Unmetered.

Rename or re-scopeLive

artifact.update

Change a name or flip public/private.

Rename an artifact, or switch its visibility. Turning a public site private mints a share token so you keep a working link.

Rotate share tokenLive

artifact.rotate_token

Mint a new share link and kill every old one.

Issue a fresh share token for a private artifact. Every link handed out before this call stops working immediately.

Delete an artifactLive

artifact.delete

Take a site down; the URL stops resolving immediately.

Remove a deployed artifact and everything it served. Unmetered, and not reversible.

Skill packs

6· Curated bundles we're building next, each chaining tools you already have.

Social Graphics KitComing soon

One brief in, a full set of on-brand post images out.

Describe the post once and get every asset it needs — sized and restyled per platform, then published to a link your team can review.

Composes

image.generateimage.editartifact.deploy

Deck BuilderComing soon

Draft a slide deck with generated visuals, publish it as a link.

Turn an outline into a full deck: generated imagery per slide, consistent styling, and a shareable URL instead of a file attachment.

Composes

image.generateartifact.deploymemory.get

Meeting RecapComing soon

Drop in a recording; get a decisions-and-owners page.

Transcribe the call, pull out decisions and action items, remember the running context, and publish a recap page the team can open.

Composes

audio.transcribememory.saveartifact.deploy

Research BriefComing soon

Search the live web, read what it turns up, publish a cited brief.

Sweep current sources, read the charts and screenshots they contain, and assemble a brief with citations you can hand to someone else.

Composes

search.webimage.describeartifact.deploy

Voice LoopComing soon

Talk to your agent and have it talk back.

A full spoken round trip: it listens, answers out loud, and remembers what you covered so the next call picks up where the last one ended.

Composes

audio.transcribeaudio.synthesizememory.savememory.search

Screenshot to SpecComing soon

Hand it a UI screenshot; get a spec you can hand to an engineer.

Read a design or a competitor's screen, write out the components, states, and copy, and publish the spec as a page instead of a wall of chat.

Composes

image.describememory.saveartifact.deploy

Web search and artifact publishing are advertised over MCP when they are configured on the deployment — both are on for router9.com. See the MCP tool reference for the full schemas.

Set up MCP in one minute

Pick your client and paste the config. Router9 hosts the server — nothing to install.

https://api.router9.com/v1/mcp

.mcp.json (project) · ~/.claude.json (user)

{
  "mcpServers": {
    "router9": {
      "type": "http",
      "url": "https://api.router9.com/v1/mcp",
      "headers": {
        "Authorization": "Bearer ${ROUTER9_API_KEY}"
      }
    }
  }
}

Claude Code expands environment variables, so the key never has to be committed. A literal sk-r9k-... works too.

CLI setup and per-client details

Works with your harness

Router9 speaks the OpenAI API, so any agent runs on it as its model — then add the Skills over MCP (or call the REST endpoints directly). Pick yours for a setup guide.

Coding agents

Agents & CLIs

OpenWorkREST endpoints

NanoClawREST endpoints

MCP tools = Skills exposed as native agent tools · REST endpoints = call them directly. Claude Code connects for the tools over MCP and keeps its own model provider. Using something else? See all integrations.