Skip to content
Developer Codex
Concepts

AI in Inklings

AI in Inklings operates on a clear principle: your workspace, your data, your keys. There are no background AI features running without your knowledge, no content being processed on remote servers without your configuration, and no AI features required to use Inklings productively. AI is opt-in, at every layer.

This article explains how the two AI integration paths work — the built-in agent and the MCP server — and where semantic search fits into the picture.

Inklings provides two ways to connect AI to your workspace:

The built-in agent Pro+Agent is a conversation interface embedded in Inklings. You send messages; the agent responds. Behind the scenes, the agent has access to your workspace through a set of tools — it can search your pages, read page content, create new pages, and more. You bring your own API key (BYOK) for an LLM provider; Inklings provides the execution environment, tools, and workspace context.

The MCP server Pro exposes your workspace to external AI tools through the Model Context Protocol (MCP). Any MCP-compatible AI tool — Claude, Cursor, or another MCP client — can connect to your Inklings workspace over a local HTTP connection, using the same set of tools available to the built-in agent. You configure the connection once; the external tool does the rest.

These two paths serve different workflows. The built-in agent is for in-context creative work: brainstorming, questions about your world, help with continuity problems. The MCP server is for integrating Inklings with tools you already use, or for using an AI client you prefer.

Pro+Agent

The agent operates as a conversation within Inklings. Open the agent panel, type a message, and the agent responds — using your workspace as its source of truth.

The agent’s capabilities come from its tool set. Tools let the agent take actions in your workspace:

ToolWhat it does
search_pagesFull-text search across your workspace
read_pageReads a page’s content and metadata
get_page_treeReturns the workspace hierarchy
get_backlinksLists pages that link to a target page
create_pageCreates a new page
update_page_contentEdits a page’s content
move_pageMoves a page in the hierarchy
rename_pageRenames a page (propagates to all links)
delete_pageMoves a page to trash

The agent can chain tool calls across multiple turns. Ask it to “read all the chapter notes and identify subplots that dropped off” — it will search, read, and synthesize across multiple pages in a single response.

Behind the conversation interface, the agent operates as a team. The Orchestrator is the agent you talk to — it coordinates specialists who handle specific tasks. A Researcher investigates your workspace (read-only), a Worker executes changes, and an Archivist manages what the agent remembers across sessions. You interact with the Orchestrator; the team works behind the scenes.

The agent maintains persistent memory across conversations. Facts about your world, your preferences, and patterns it notices are stored locally and retrieved when relevant. See Agent Memory.

You provide the API key. Navigate to Settings → Agent → Configure Provider, select your LLM provider (Anthropic, OpenAI, xAI, Ollama, or OpenRouter), and enter your key. Inklings stores the key in your OS keychain — it’s not stored in the database or transmitted to Inklings servers.

When the agent runs, your API key is used to make requests to the provider you’ve configured. Inklings assembles the requests (system prompt, conversation history, tool schema, workspace context) and routes them to your provider. The LLM response comes back to Inklings; the agent processes it and executes any tool calls.

This means: no Inklings-managed AI subscription, no opaque model selection, and no data going through an Inklings relay. The request goes from your machine to your chosen provider, directly.

The agent operates within the same permission model as everything else in Inklings. Tools that require write access (create, update, delete) are gated by the PagesWrite capability. Tools that read are gated by PagesRead. When you run as the workspace owner — which is the normal case for local workspaces — you have all capabilities.

The agent’s tool schema shows which tools are available and which are unavailable given the current permission context. Unavailable tools appear with a note in the schema; if the agent attempts to call one, the call is rejected before it reaches the workspace.

Pro

The MCP server runs inside Inklings as a local HTTP service, bound to 127.0.0.1 on a configurable port (default 7862). No external network access is required; the server is only reachable from your own machine.

Enable the MCP server from Settings → Integrations → MCP Server. When enabled, a bearer token is generated — a 64-character cryptographic key that protects the /mcp endpoint. Copy the token and the connection URL from settings, then configure your MCP client with both.

The MCP server exposes:

  • The same workspace tools available to the built-in agent
  • A workspace tree resource (inklings://workspace/tree)
  • Page resources (inklings://workspace/page/<slug>)
  • A search resource (inklings://workspace/search?q=<query>)

External tools connect, authenticate, and then have access to your workspace through the MCP protocol. From the tool’s perspective, Inklings is a workspace server. From your perspective, your external AI tool can now read and write your Inklings pages.

The MCP server starts automatically when you open a workspace (if it was enabled when you last closed it) and stops when you close the workspace or quit the app.

Semantic search — finding pages by meaning rather than exact words — is built into Inklings and requires no AI provider configuration. It runs locally using a built-in AI model.

When you search your workspace, results combine full-text search (exact and prefix word matches) with meaning-based matching. The combination is ranked using a scoring algorithm that weights title matches highest, then tag matches, then content matches.

Semantic search doesn’t send your content anywhere. The embedding model runs on your machine, processes your pages locally, and stores the results in your workspace database. No API key, no cloud service.

FeatureTier
All pages, editor, hierarchy, tags, types, searchFree
Semantic searchFree
MCP serverPro
Built-in agent (conversation, tools)Pro+Agent
Configure LLM providersPro+Agent
Skills marketplacePro+Agent

The Free tier is fully functional for all PKM and creative writing workflows. AI features are additive.

Was this page helpful?