Use cases / Shared agent context

Switch agents. Keep your knowledge.

Give compatible AI tools access to the same local wiki instead of rebuilding context in every chat.

Keep one durable wiki

Record accepted decisions, terminology, and project constraints in Markdown. Every session can consult the same files, including when you switch coding tools.

Use agent instructions to tell the agent when to consult and update the wiki.
Search for the relevant sections instead of attaching the whole wiki to every request.
Keep ephemeral task state separate from accepted project knowledge.

Connect a compatible MCP client

Client MCP configuration

{
  "mcpServers": {
    "project-knowledge": {
      "command": "okn",
      "args": ["mcp", "/absolute/path/to/Wiki"]
    }
  }
}

Replace the absolute path with your wiki location. Configure each client separately. Local MCP is read-only: it exposes search and source access, not automatic chat-memory capture.

Browse and edit together with WebMCP MCP configuration reference

Retrieve a focused context packet

okn search Wiki "release workflow" --budget 1200

Example output · repository wiki excerpt

Context: 1195 / 1200 estimated tokens
Sources: 6

Source: `workflows/changelog-updates.md:30-53`

Add an entry under `## Unreleased` in the
[CLI changelog](/changelog/cli.md).

A bounded context packet keeps source references. Your agent uses it to produce an answer.

The result includes source paths, line ranges, and selected text within an approximate token budget. It is retrieved context; your agent produces the answer.

Agents update the Markdown through their normal file tools or the local editor. Save and review those changes explicitly.

See the knowledge it uses

This repository’s own wiki documents its commands, design decisions, and release history.

Open the Open Knowledge wiki

Set up your knowledge base

Follow the setup guide Explore other workflows