Recipe: Chat With Your Blog

Your friend Dave has a decade of blog posts on WordPress. He wants to ask his AI assistant questions like “What did I write about microservices?” and get answers sourced from his actual posts — not the internet, not hallucinations, just Dave’s own words.

Here’s how to set that up in five minutes.

The idea

1. A Lua connector parses Dave’s WordPress XML export into individual posts (see Recipe: Index Your WordPress Blog)
2. An inline agent defines a persona that answers as Dave using only his writing
3. Cursor (or Claude Desktop, or any MCP client) connects to the agent and Dave can chat with his own blog

The config

One file does everything — config/ctx.toml:

[db]
path = "./data/ctx.sqlite"

[chunking]
max_tokens = 700
overlap_tokens = 80

[retrieval]
final_limit = 10
hybrid_alpha = 0.6

[server]
bind = "127.0.0.1:7331"

# ── Connector: ingest Dave's blog ────────────────────────────
[connectors.script.wordpress]
path = "connectors/wordpress.lua"
file = "wordpress-export.xml"

# ── Agent: chat with Dave ────────────────────────────────────
[agents.inline.dave]
description = "Chat with Dave — answers based on his published blog posts"
tools = ["search", "get"]
system_prompt = """
You are Dave — a conversational persona grounded entirely in Dave's
published blog posts. When answering questions:

1. Use the `search` tool to find relevant posts
2. Use the `get` tool to read full post content when snippets aren't enough
3. Quote or paraphrase Dave's actual writing — cite post titles
4. If Dave never wrote about a topic, say so honestly
5. Match Dave's writing style — casual, opinionated, practical

Never invent facts. Never attribute ideas to Dave that aren't in his posts.
You ARE Dave, speaking from your own published work.
"""

Set it up

# Put the WordPress export next to the connector script
cp ~/Downloads/wordpress-export.xml connectors/

# Initialize, sync, and optionally embed
ctx init
ctx sync script:wordpress
ctx embed pending   # optional: enables semantic search

# Start the server
ctx serve mcp

Connect Cursor

.cursor/mcp.json:

{
  "mcpServers": {
    "dave": {
      "url": "http://127.0.0.1:7331/mcp"
    }
  }
}

Open Cursor and select the dave agent. Now you can ask:

• “What’s your take on microservices?”
• “Did you ever write about database migrations?”
• “Summarize your Kubernetes series”
• “What changed in your thinking about testing between 2022 and 2024?”

Every answer is grounded in Dave’s actual posts, with citations.

Make it smarter with a Lua agent

The inline TOML agent above is static — it always uses the same system prompt. If you want the agent to pre-fetch relevant posts before the conversation starts (RAG priming), use a Lua agent instead:

[agents.script.dave]
path = "agents/dave.lua"
timeout = 30
search_limit = 5

-- agents/dave.lua
agent = {}
agent.name = "dave"
agent.description = "Chat with Dave — grounded in his published blog posts"
agent.tools = { "search", "get" }

agent.arguments = {
    {
        name = "topic",
        description = "Optional topic to pre-load relevant posts",
        required = false,
    },
}

function agent.resolve(args, config, context)
    local topic = args.topic or "recent writing"

    -- Pre-search for relevant posts
    local results = context.search({
        query = topic,
        mode = "hybrid",
        limit = config.search_limit or 5,
    })

    -- Build context from top results
    local context_text = ""
    for _, r in ipairs(results) do
        local doc = context.get(r.id)
        context_text = context_text .. "\n\n---\n## " ..
            (doc.title or "Untitled") .. "\n" .. doc.body
    end

    return {
        system = string.format([[
You are Dave. You answer questions based on your blog posts.

Here are your most relevant posts on "%s":
%s

Use the search and get tools to find more posts if needed.
Quote your own writing. Cite post titles. Be yourself.
        ]], topic, context_text),

        messages = {
            {
                role = "assistant",
                content = string.format(
                    "I've loaded %d of my posts about %s. What do you want to know?",
                    #results, topic
                ),
            },
        },
    }
end

return agent

Now when someone selects the dave agent with topic=kubernetes, the agent pre-searches for Kubernetes posts and injects them into the system prompt before the conversation even starts.

Beyond WordPress

This pattern works for any personal writing:

The connector parses the source format. The agent defines the persona. Context Harness handles everything in between.