Eclipse MCP

Connect Claude Code, Cursor, Codex, or Claude Desktop to Eclipse and run content, research, GEO, social and client work from your agent.

Eclipse MCP

Eclipse MCP lets an external AI agent, such as Claude Code or Cursor, operate Eclipse for you. It is separate from CASi, the AI built into Eclipse. Create a token in settings, paste a short config into your agent, and ask it to run the work in the sections below. Draft an article, pull last month's performance, run competitor research, schedule a week of social posts.

MCP (Model Context Protocol) is the open standard AI tools use to connect to products like Eclipse. There is nothing to install. Claude Code, Cursor and Codex connect over HTTPS with your token. For Claude Desktop, the connection is a one-line bridge.

Agencies can use one token across their assigned clients, or bind a token to a single client.

What your agent can do

Ask in plain language, the way you would brief a colleague. The agent uses the same client data you see in Eclipse, shows a preview before it publishes or spends, and checks back when a job takes a few minutes.

Content and Planner

Pull the content table, open an article, and change the title, body, excerpt or metadata. Comments land on the piece and stay attributed to you. You can also count articles by status or type inside a date window, and manage authors.

List the Content Profiles on the client, or create a draft from a product, ideal customer profile, audience and industry already saved there. Those records supply the names and descriptions. Publish the draft once its keywords are focused, or send it back to draft, which clears the weekly schedule. Article writing is its own step. Medium profiles need the seven editorial questions answered before publishing. On a published profile, set the days of the week it should generate, Sunday through Saturday, or clear that list. Scheduled generation spends credits on the account you are acting as, under the usual billing rules. Delete a profile that has no generated articles and no generation in progress. Failed generation attempts do not block deletion.

New articles start as a brief. Set the topic, title, points to cover, and any research or instructions, then the agent writes the draft. Writing uses one article credit. On the Planner, update the topic, title or approval status, archive a row, or set the time Eclipse should write it. A title suggested from a Scratch Pad note can land on the Planner as a draft row.

When WordPress, Webflow, HubSpot or Contentful is already connected and the fields are mapped, you can publish the article live or save it as a draft, and a later publish to the same destination updates the item already there. You can also check whether a piece is live, a draft or scheduled, and whether the destination copy is behind Eclipse.

Try "Draft a brief on our onboarding guide, write it, and tell me which CMS destinations can take it." Ask "Create a draft profile, then publish it once the keywords are focused." Writing Content, Planner and Content Profiles walk through the same jobs in the web app.

Scratch Pad and Library

Capture a rough thought, tag it, and link it to the notes already in that client. Those notes stay private to the person who owns the token. Library items follow that same privacy rule inside the client. From a note you can generate blog title suggestions, send the note onto a Cortex canvas, or save canvas work into the Library.

Try "Capture these three points as a Scratch Pad note, suggest titles, and add the strongest one to the Planner." Scratch Pad covers the same notes in the web app.

Cortex and CASi

Open a canvas or create one. Add text, a URL, a YouTube link, a competitor or a library item. You can also add a sticky note, a group, a chat card, an article, a social post, a monitor result, a Reddit thread, a podcast, a document, a voice memo or an image. Connect the cards that should feed each other, or disconnect a link. Chat with CASi on that canvas and run deep research or community research the same way you would on the canvas. When the findings are in, generate an article from the nodes you choose, or train CASi on a node so later writing can use it.

On an image card, write the prompt yourself or generate a brief from the cards you connected. Pick a size and a mode (Fast, Standard or Pro), turn the brand theme on or off, and add reference images if you want them. Generate the visual or edit a version, then attach it to an article or a social post, or detach it later.

Chat, research, image briefs, image generation and article generation spend AI balance or credits, and the preview shows that cost before anything starts.

Try "Open the pricing research canvas, run deep research on the objection we keep hearing, and draft an article from those nodes." Ask "Brief an image from those cards, generate it, and attach it to the draft." Cortex Canvas, CASi, Deep Research and Image Generation cover this work in the browser.

Train CASi

Upload text, a web page or a file, then file it in a folder so CASi can draw on it later. You can list, update, move and delete training documents, and check whether a canvas node is trained, out of date or still processing. Each client's material stays in that workspace.

Try "Upload this product sheet to Train CASi and put it in the sales folder." Train CASi explains what to upload and how CASi uses it.

GEO

Read the visibility overview, the platform breakdown, trends, tracking, history and insights. Create a profile as a draft, publish it once the prompt and platforms are set, and start an analysis run that charges the client AI balance while the agent follows the job.

Changing the prompt or platforms on a profile that already exists stays in the web app.

Try "Read our AI search visibility and start a new run if the category profile is stale." GEO Dashboard and GEO Profiles show the same numbers in the web app.

Keywords

Read rank, movement, search volume and difficulty for the keywords you track, including focus keywords. Add a keyword for a region, remove one, read daily rank history, or queue a refresh of current ranks.

Try "Show focus keywords that moved over the last 30 days."

Keywords shows the same table in the web app.

Competitors

Read the profile Eclipse already holds for a rival, including the stored evidence. You can start monitoring a competitor Eclipse has already found, add any other domain and start research, refresh one profile now, or schedule a removal for the end of the current billing cycle. Each of those waits for a preview before it runs, and adding a competitor shows the monthly slot price.

Try "Summarise what we hold on example.com and refresh it if the profile is stale." Competitor Intelligence is the same view in the web app.

Monitors

List the monitors on a client, read what they found, and pause, resume or archive one. Create a monitor for web, news, global media, trusted sources, or a mix, on an hourly, daily or weekly check. The agent confirms before it saves the monitor. Creation itself is free. Each later check is charged to AI balance, and you can set a monthly spend cap when you create it.

Try "Show what the funding monitor found this week." Monitors Overview explains sources, schedules and alerts.

Analytics

Read Google Analytics traffic and Google Search Console performance when those accounts are already connected. You can look at the last 7, 30, 90 or 365 days, with the previous period shown beside the traffic figures. The agent can also tell you which property and site are selected. Connecting the accounts stays in Eclipse.

Try "Compare organic traffic with the previous 30 days." Analytics, Google Analytics and Google Search Console cover how those accounts are connected.

Start with the subreddits you already monitor. List open opportunities, brand and competitor mentions, and inbox messages, then draft a reply, refine the wording, and post it to the thread when you are ready. You can also read engagement, growth guidance and Reddit referral traffic. Drafting spends AI balance, posting waits for a preview, and the Reddit account itself is connected in Eclipse.

Try "Show open Reddit opportunities and draft a reply to the strongest one, and leave it unposted." Reddit Engagement, Inbox, Activity, Growth, and Analytics cover that work in the browser.

List social profiles or create one, then draft posts from an article, a Scratch Pad note or news stories. Edit the draft text, publish now, or schedule a time. Destinations are the Buffer channels or direct connections already set up for LinkedIn, X, Facebook, Instagram, Pinterest, Threads, TikTok or YouTube. Each publish or schedule call uses one connection type. Drafting spends credits per post and can take up to ten minutes, while a post that is already scheduled or published is edited in Eclipse.

Try "Turn yesterday's article into three LinkedIn drafts and leave them for review." Social Content and Direct Social Publishing cover profiles, Buffer and direct connections.

Napkin visuals

Scan an article or a social post for passages that suit a visual. Pick one or dismiss it. The agent reads the client's default style and custom brand, and can change them for later visuals. From the passage you choose, it generates variants in a style you pick, follows that job until it finishes, and you select the visual you want to keep. The visual you select lands in an article, while a social post takes the first finished visual on its own. List or download the visuals on a social post. Abandon a generation that is still running.

Napkin generation returns an operation id, and the agent checks get_operation until the status is done or failed. These Napkin steps are free.

Try "Find a passage in yesterday's article for a Napkin visual and generate one in our default style." Napkin AI covers styles, variants and brand setup in the editor.

Clients, brand and agency

Read and update client details, and generate a brand description, tone of voice or author bio the same way Ask CASi does in the app. Add products, audiences, industries and ideal customer profiles so the next draft has a defined reader.

Agency owners can list every client the token can see, create one from a name, website and tracking region, and archive or restore a client. They can also list staff and assign or unassign someone, and Eclipse emails a person when they are assigned. A token only reaches the clients it was granted.

Try "List my clients and draft a tone of voice for the one still missing it." Client Management and Agency Accounts explain workspaces and team access.

Billing

Check article credits and AI balance, review the transactions on the account, and see which packs you can buy, then confirm a purchase so Eclipse adds the balance straight away and issues an invoice for the pack. Agency staff who cannot purchase can request a top-up, and Eclipse emails the admin.

While a trial has ended and billing details are still missing, the account payer can submit them from the agent, and agency staff cannot. Changing the plan stays in the web app. The balance pages live under Credits & AI Balance.

Create a token

Go to Settings and open the AI agent section.
Create a token. Give it a name, choose an expiry if you want one, and choose which clients it can access.
Copy the token straight away. Eclipse shows it once.

You can list and revoke tokens from the same panel at any time. Revoking a token cuts the agent off on its next request.

Set up your agent

Pick your agent below and replace YOUR_ECLIPSE_TOKEN with the token you just created. Restart the agent, then ask it something. "What did we publish last month?" is a good first test.

Claude Code

Run this once in your terminal.

claude mcp add --transport http eclipse https://app.geteclipse.co/api/mcp --header "Authorization: Bearer YOUR_ECLIPSE_TOKEN"

Cursor

Add this to ~/.cursor/mcp.json, then reload Cursor.

{
  "mcpServers": {
    "eclipse": {
      "url": "https://app.geteclipse.co/api/mcp",
      "headers": {
        "Authorization": "Bearer YOUR_ECLIPSE_TOKEN"
      }
    }
  }
}

Codex

Export the token in your shell profile.

export ECLIPSE_MCP_TOKEN="YOUR_ECLIPSE_TOKEN"

Then add the server to ~/.codex/config.toml.

[mcp_servers.eclipse]
url = "https://app.geteclipse.co/api/mcp"
bearer_token_env_var = "ECLIPSE_MCP_TOKEN"

Claude Desktop

Claude Desktop connects through the mcp-remote bridge. Add this to claude_desktop_config.json, then restart Claude Desktop.

{
  "mcpServers": {
    "eclipse": {
      "command": "npx",
      "args": [
        "-y",
        "mcp-remote",
        "https://app.geteclipse.co/api/mcp",
        "--header",
        "Authorization:${AUTH_HEADER}"
      ],
      "env": {
        "AUTH_HEADER": "Bearer YOUR_ECLIPSE_TOKEN"
      }
    }
  }
}

Permissions, billing and audit

Your agent works inside the same boundaries as your account.

Access matches the web app, so a token bound to one client cannot reach other clients. An agency token that covers several clients names the client on each request. Eclipse never hands credentials or secrets to the agent.
Reading data is free, and a preview comes back before the agent publishes an article or social post, posts a Reddit reply, starts paid research or generation, adds or refreshes a competitor, schedules a competitor removal, or buys credits or AI balance. The action runs only after the agent confirms.
Those actions take an idempotency key, a fresh id the agent creates for that single intention. The agent sends the same key if the call drops. Eclipse replays the completed result, so a publish, a charge or a purchase happens once. Creating a client uses the same kind of key, so a retry returns the workspace you already added, with the same name, domain and tracking region.
Actions are logged against the token, and any spend appears in the client ledger as usual.
Article generation, social drafting, deep research, community research, image briefs, image generation, GEO analysis, competitor refresh, and brand, tone or author-bio generation return an operation id, and the agent checks get_operation until the status is done or failed. It then reads the article, the post, the image, the brief or the findings, and bigger drafting jobs can take up to ten minutes.

What stays in the web app

A few jobs still need a person in Eclipse.

Connecting or disconnecting integrations, including CMS accounts, Buffer, direct social profiles, Google Analytics, Search Console, Reddit and Slack. The agent can list what is already connected.
Renaming a Content Profile, or changing its targeting, keywords, article type or Medium editorial questions after it exists, stays in the web app.
Changing a GEO profile that already exists stays in the web app, while creating a draft, publishing it and running analysis can happen from the agent.
A social post that is already scheduled or published is edited in Eclipse.
Moving cards around on the canvas by hand stays in the web app.
The agent can add nodes, connect or disconnect them, run research and generate from them, and brief, generate and attach canvas images.
Subscription and plan changes, user accounts, passwords and roles stay in the web app, as does allocating agency credits between clients. Staff can request a top-up, and an admin completes the purchase.

Areas at a glance

Your agent already sees every live tool as soon as it connects, so this list is only a map of the areas. Start with the area you care about.

Content and Planner covers articles, briefs, comments, authors, Planner rows, CMS publish, and Content Profiles you can create, publish, schedule and delete when unused.
Scratch Pad and Library covers private notes, tags, links, title suggestions, sending a note to a canvas, and items saved in the Library.
Cortex and CASi covers canvases, connecting and disconnecting nodes, chat, research, articles generated from nodes, and canvas images you can brief, generate and attach.
Train CASi covers documents, folders, uploads and training taken from a canvas node.
GEO covers draft profiles, publishing them, analysis runs, and the overview, platforms, trends, history and insights you read afterwards.
Keywords covers the tracked list, rank history and a refresh of current ranks.
Competitors covers stored profiles, adding a domain, a refresh and removal scheduled for the end of the billing cycle.
Monitors covers creating a watch, reading results, and pause, resume or archive.
Analytics covers Google Analytics traffic and Google Search Console performance for a property you have already connected in Eclipse.
Reddit covers subreddits, opportunities, mentions, inbox, drafts, published replies, engagement and growth.
Social covers profiles and drafts, plus publishing or scheduling onto Buffer or a direct connection you have already set up.
Napkin visuals covers passage suggestions, styles and brand defaults, generation, choosing a variant, downloads and abandoning a run that is still going.
Clients and agency covers client details, brand lists, new clients and assigning staff.
Billing covers balances, packs, purchases and a top-up request to an agency admin.
Integrations is a read-only view of what is already connected, with the connect and disconnect steps left in Eclipse.

Was this page helpful?