agent-first architecture

Ideas saved. Now agent-ready.

A workspace your agents can read from, write to, and build on. Plug in Claude, Cursor, or anything that speaks MCP — and stop losing the good parts of a conversation.

Works with
Claude
Claude Code
Cursor
ChatGPT

Ideas die in three places.

01chat history

Useful thinking evaporates the second the session ends.

02flat markdown

Saved to notes, but with no structure for tasks, links, or status.

03one-time decks

Captured for one audience, then gone.

We gave Idealist an MCP server so the conversation itself becomes structured input.

// the workspace on the other end

Meet Idealist.

Idealist is where ideas go to grow up. Capture a thought fast. Develop it with notes and research. Break it into tasks. Track it from exploring to shipped.

It's the space between a notes app (too loose) and a project manager (too heavy). The web app, a Chrome extension, a command palette, and an AI coach all point at the same workspace. MCP is one more way in.

Usage-based pricing
Exploring
3 notes · 1 taskai-drafted
Redesign portfolio
Planning
8 tasks · 5 notes
Launch blog
Executing
2/4 tasks
SaaS landing page
Shipped
12/12 tasks · shipped
// three agent workflows

What agents actually do with it.

01save
workflow

Save it from the chat.

End-of-conversation rescue. The agent packages up what you just discussed as a structured idea with notes, tasks, and links.

you >save the Berlin launch plan as an idea
create_idea{ title: "Berlin launch" }
add_note{ content: "Venue, team, timing" }
add_task{ title: "Confirm venue" }
✓ 1 idea · 1 note · 1 task
02pull
workflow

Pull it back into the chat.

Context on demand. The agent fetches the idea and all its notes — so next week's conversation starts where last week's left off.

you >what was our Berlin launch plan?
search_ideas{ query: "Berlin" }
get_idea{ id: "idl_9b2e" }
claude >Berlin launch · planning · 3 notes · 2 tasks.
03extend
workflow

Extend what's already there.

Additive, not destructive. Agents append new notes, tasks, and links. Existing content stays exactly as you wrote it.

you >add a task: book travel by Friday
add_task{ title: "Book travel", priority: "high" }
✓ task added · tagged ai-drafted
// about 30 seconds

Connect it.

Paste the endpoint and a token into your MCP client. Reads stream. Writes are rate-limited to 100/hour/token.

~/claude_desktop_config.json
{
  "mcpServers": {
    "idealist": {
      "url": "https://idealist-app.com/api/mcp",
      "headers": {
        "Authorization": "Bearer idl_your_token_here"
      }
    }
  }
}

Tokens live in Settings → MCP once access is granted. Hashed on the server, shown once, revocable any time.

// the full surface

Seven tools to manage your ideas.

A small, opinionated toolset. Writes-only in v1 — edits and deletes stay your call.

  • list_ideas
    read
    in:{}
    out:{ ideas[] }
    Lightweight index: id, title, status, updated_at.
  • get_idea
    read
    in:{ idea_id }
    out:{ idea, notes[], tasks[], links[] }
    The full object. Notes return full markdown.
  • search_ideas
    read
    in:{ query, limit? }
    out:{ results[], truncated }
    Fuzzy match across titles, descriptions, and note contents.
  • create_idea
    write
    in:{ title, description?, status?, tags? }
    out:{ idea }
    Auto-tagged 'ai-drafted'. Attribution logged.
  • add_note
    write
    in:{ idea_id, title?, content }
    out:{ note }
    No imposed structure. The agent uses judgement.
  • add_task
    write
    in:{ idea_id, title, priority? }
    out:{ task }
    Section-less in v1. Priority and completed supported.
  • add_link
    write
    in:{ idea_id, url, title? }
    out:{ link }
    OpenGraph metadata fetched automatically.

7 tools · 3 read · 4 write

// frequently asked

Questions worth answering.

Getting started
How it works
Security & limits
Roadmap
// pilot access

Get on the list.

MCP is enabled for a small group while we iterate on tool ergonomics. Drop your email and we'll reach out as slots open.