wikis / Agent Setup

Agent Setup

Connect AI agents to Pixi Wiki so they can list, search, and read the same Markdown knowledge bases behind the human wiki.

Why

Subagents do not inherit your full context, preferences, voice, project priorities, or thesis automatically. Make retrieval explicit before research or writing.

Access

Read-only local MCP over index.json and raw/<kb>/**/*.md. No write, edit, or delete tools.

Use

Good for project context, public knowledge, portfolio framing, source-backed retrieval, and agent onboarding.

Connect the MCP server

git clone https://github.com/pixiiidust/pixi-wiki.git
cd pixi-wiki
python3 -m pip install mcp
python3 scripts/pixi_wiki_mcp.py --self-test

Then configure your MCP client to launch:

python3 /path/to/pixi-wiki/scripts/pixi_wiki_mcp.py

Hermes config

mcp_servers:
  pixi_wiki:
    command: "python3"
    args: ["/root/pixi-wiki/scripts/pixi_wiki_mcp.py"]

Restart Hermes after editing the config. Tools appear with the mcp_pixi_wiki_ prefix.

Recommended agent workflow

Call list_kbs.
Call get_kb_summary for relevant KBs.
Call search_all_kbs for the task topic.
Call read_document for the best matches.
Tailor the answer to retrieved context instead of producing generic research.

Subagent instruction template

Before doing research, writing, product strategy, or project analysis, use the Pixi Wiki MCP server. Start with list_kbs, then search_all_kbs for the topic, then read the most relevant documents. Tailor the answer to Jamie's documented project context and preferences. Do not produce generic output if Pixi Wiki has relevant context.

Important boundary

Pixi Wiki MCP exposes public compiled KBs. Private memory, live chat context, secrets, and profile-only preferences are separate unless intentionally compiled into a private/local-only KB.

Full MCP server reference →