Documentation

Everything you need to connect CEPO to your AI tools.

What is CEPO

CEPO is a personal AI workspace that gives your AI agents persistent memory and a way to ask you questions when they need human input.

It works through the Model Context Protocol (MCP), which means it integrates with any AI tool that supports MCP — including Claude Desktop, Claude Code, Cursor, and Windsurf.

Custom MCP Servers

CEPO lets you create multiple custom MCP servers, each with a different set of tools. This means you can give each AI agent exactly the capabilities it needs — no more, no less.

Each custom MCP server gets its own OAuth credentials and URL. You choose which apps (and their tools) are included when you create it.

Example configurations

“research-tools” — for a research agent

Apps: Knowledge Base + Tags. The agent can save findings, search notes, and organize with tags — but can't create HITL requests.

“review-tools” — for a code review agent

Apps: HITL + Knowledge Base. The agent can ask you approval questions and save review findings — scoped to just what a reviewer needs.

“full-access” — for a general-purpose assistant

Apps: All. The agent has access to every tool — useful for a primary assistant that handles everything.

Available Tools

Tools are grouped by app. When you create a custom MCP server, you select which apps to include — all tools from those apps become available to the connected agent. Guide tools are added automatically.

Knowledge Base

Persistent notes that your AI agents can read and write.

write
kb_create

Create a new knowledge base note

read
kb_get

Get a knowledge base note by ID

write
kb_update

Update an existing knowledge base note

read
kb_search

Search knowledge base notes by query and optional tags

read
kb_list

List knowledge base notes with optional pagination and tag filtering

Tags

Organize notes and resources with custom tags.

read
tags_list

List available tags for categorizing and filtering resources

write
tags_create

Create a new tag for categorizing and filtering resources

Prompts

Saved prompts that your AI agents can browse and retrieve.

read
prompts_list

List saved prompts with optional filtering

read
prompts_get

Get a specific prompt by ID

Human-in-the-Loop (HITL)

Let agents ask you questions and wait for your response.

write
hitl_create

Create a human-in-the-loop request for user input

read
hitl_get

Poll the status of a human-in-the-loop request

Guides

Every custom MCP also exposes read-only guide tools. guide_how_to_use is always included, and one guide is added for each app you select — the agent reads them to learn how the apps work.

read
guide_how_to_use

What Cepo.ai is and which guide to read next. Start here before using the other tools.

read
guide_kb

How to use the knowledge base: note hierarchy, doc vs ticket, tags, and the Markdown formats that render.

read
guide_tasks

How to use tasks: priority, due dates, subtasks, completion, and filtering by tag.

read
guide_projects

How to use projects to group tasks and notes, and how work attaches to one.

read
guide_tags

How tags work across apps: names, hierarchy, filtering, and the MCP visibility trap.

read
guide_prompts

How to find the user's saved prompts and interpolate their variables.

read
guide_hitl

How to ask the user a question and poll for their answer.

Setup

After creating a custom MCP server in your MCP Hub, you'll get a server URL. Add it to your AI tool's MCP configuration.

Claude Desktop

Add to your Claude Desktop MCP settings:

{
  "mcpServers": {
    "cepo": {
      "url": "https://cepo.ai/my-mcps/<your-mcp-slug>"
    }
  }
}

Claude Code

Add using the CLI:

claude mcp add cepo \
  --transport http \
  https://cepo.ai/my-mcps/<your-mcp-slug>

Cursor

Add to your .cursor/mcp.json:

{
  "mcpServers": {
    "cepo": {
      "url": "https://cepo.ai/my-mcps/<your-mcp-slug>"
    }
  }
}

Authentication

CEPO uses OAuth 2.1 with PKCE for authentication. When you first connect an AI tool to your MCP server, you'll be redirected to sign in with your CEPO account and authorize the connection.

Each custom MCP server has its own OAuth client with scoped permissions. Tokens are automatically refreshed — you only need to authorize once per MCP server.

Usage Examples

Here are real-world examples of how agents use CEPO tools.

1. Save a debugging insight

MCP: research-tools (Knowledge Base + Tags)

“Save this to my knowledge base: We fixed the N+1 query issue in the user dashboard by eager-loading associations. Tag it with 'performance' and 'database'.”

The agent calls kb_create with the title, content, and tags. The note is saved to your knowledge base and tagged for future retrieval.

2. Search and organize notes with tags

MCP: research-tools (Knowledge Base + Tags)

“What debugging notes do I have about Cloudflare Workers? Create a 'cloudflare' tag and add it to any relevant notes.”

The agent calls kb_search to find matching notes, then tags_create to create the new tag, and finally kb_update on each relevant note to associate the tag.

3. Ask for human approval before proceeding

MCP: review-tools (HITL + Knowledge Base)

“Ask the developer if we should proceed with the database migration. Include the migration summary for context.”

The agent calls hitl_create with an approval question. You receive a notification in your CEPO inbox. The agent then polls with hitl_get until you respond — approving, rejecting, or providing additional input.

4. Build context from past decisions

MCP: full-access (All apps)

“Before we start the auth refactor, check my knowledge base for any past decisions about authentication patterns.”

The agent calls kb_search with relevant keywords, then tags_list to see if there are related tags. Results inform the agent's approach — your past decisions become institutional knowledge that persists across sessions.

Privacy & Support