Skip to content
edisyledisyl

Search these docs from your AI tool

  • Antigravity
    {
  "mcpServers": {
    "edisyl-docs": {
      "serverUrl": "https://docs.edisyl.com/mcp"
    }
  }
}
  • Claude CodeCLI
    claude mcp add edisyl-docs https://docs.edisyl.com/mcp --transport http --scope user
  • CodexCLI
    codex mcp add --url https://docs.edisyl.com/mcp edisyl-docs
  • Cursor
    {
  "mcpServers": {
    "edisyl-docs": {
      "type": "http",
      "url": "https://docs.edisyl.com/mcp"
    }
  }
}
  • Gemini CLI
    gemini mcp add --transport http edisyl-docs https://docs.edisyl.com/mcp
  • Goose
    {
  "extensions": {
    "edisyl-docs": {
      "enabled": true,
      "name": "edisyl-docs",
      "type": "streamable_http",
      "uri": "https://docs.edisyl.com/mcp",
      "envs": {},
      "env_keys": [],
      "headers": {},
      "description": "",
      "timeout": 300,
      "bundled": null,
      "available_tools": []
    }
  }
}
  • JetBrains AI Assistant
    {
  "mcpServers": {
    "edisyl-docs": {
      "type": "http",
      "url": "https://docs.edisyl.com/mcp"
    }
  }
}

    Paste into Settings → Tools → AI Assistant → Model Context Protocol → Add → As JSON. Direct file writing isn’t supported — JetBrains stores this per-version as XML.

  • Junie (JetBrains)
    {
  "mcpServers": {
    "edisyl-docs": {
      "url": "https://docs.edisyl.com/mcp"
    }
  }
}
  • OpenCode
    {
  "mcp": {
    "edisyl-docs": {
      "type": "remote",
      "url": "https://docs.edisyl.com/mcp"
    }
  }
}
  • VS CodeCLI
    code --add-mcp '{"name":"edisyl-docs","type":"http","url":"https://docs.edisyl.com/mcp"}'
  • Windsurf
    {
  "mcpServers": {
    "edisyl-docs": {
      "serverUrl": "https://docs.edisyl.com/mcp"
    }
  }
}

    Supports both stdio and native HTTP connections.

Looking to ask a pack from your AI tool instead? That's a different server: see AI clients.

Set up a Knowledge Pack

A Knowledge Pack puts your organization’s documents, data, and definitions in one place. Any AI tool can query it and get cited answers. This guide shows how to build a pack from the CLI, from the start until it is ready for people to use. You can run the CLI yourself or give it to a coding agent. Add documents first, then data, then the questions the pack has to answer correctly.

Before you start

Section titled “Before you start”

  • Install the CLI and log in.

    Terminal window
    curl -fsSL https://cli.edisyl.com/install.sh | sh
    edisyl profile create work --endpoint https://api.<your-domain> --use
    edisyl login --profile work
    edisyl whoami

    Use your edisyl install’s API host in place of api.<your-domain>. Your edisyl admin has it. See Install & log in.

  • Gather documents. Policies, definitions, data dictionaries, process guides. PDF, Markdown, and text all work. The more relevant documents you add, the better the answers.

  • Bring read-only warehouse credentials. A service user with read access only. For Snowflake, use key-pair auth. You enter credentials through the CLI. They never show in the app. See Connect a data source.

Let an agent drive

Section titled “Let an agent drive”

The CLI is built for AI agents such as Claude Code to run. The agent can also install and set up the CLI.

Copy this into Claude Code, Cursor, or any coding agent with terminal access. Approve the sign-in in your browser, then ask in plain language, like the prompts in each step below.

Keep the edisyl app open to see the result of each step and to change what edisyl decided.

Build the pack

Section titled “Build the pack”

Upload documents first, then connect your data. edisyl reads the documents to learn what your tables and columns mean.

Some steps also work from an AI chat client over MCP. Those steps show both ways. CLI only marks commands that have no MCP tool. You can do those steps in the app or the CLI.

  1. Create the pack. In the app, or:

    CLI only
    Terminal window
    edisyl pack create --name "Supplier risk"

    Ask your agent: Create a pack called Supplier risk.

  2. Add the people who build it. On the pack page under Who can build this, or:

    CLI only
    Terminal window
    edisyl pack members add supplier-risk ana@acme.com bo@acme.com --role contributor

    Roles are viewer (read), contributor (write), and admin. Editing PACK.md and the Lexicon needs manage_pack on the pack.

  3. Add documents. Drag files onto the pack page, or teach them from chat or the CLI. See Add documents for file types and limits.

    Attach a file in chat. The assistant calls teach with the file.

    “Teach the attached policy to the Supplier risk pack.”

  4. Check the documents. Each one shows as Indexed on the pack page.

    CLI only
    Terminal window
    edisyl pack documents list supplier-risk

    Ask your agent: Which documents are in the pack? Did any fail?

  5. Start PACK.md (optional at this stage). Write two to four sentences on the pack’s purpose, what it covers, and who it is for. See Steering with PACK.md.

    CLI only
    Terminal window
    edisyl pack context edit supplier-risk

    Ask your agent: Draft a PACK.md purpose and show it to me first.

  6. Connect your warehouse. This tests the credentials and saves them once.

    CLI only
    Terminal window
    edisyl connections create --name "Snowflake prod" --provider snowflake --config-file sf.json

    Ask your agent: Connect our Snowflake warehouse using sf.json.

  7. Add the data source to this pack. Choose the database, and optionally a schema. Pass --pack-id to add the source to this pack. Without it, edisyl creates a new pack.

    CLI only
    Terminal window
    edisyl data-sources create --name "Procurement" --connection-id <connection-id> \
      --database PROD --schema PROCUREMENT --pack-id <pack-id>

    Ask your agent: Add the PROD database to the Supplier risk pack.

    New tables that appear in the scope later are synced and enriched automatically every day. Add --no-auto-add-tables to hold them for approval instead. See Adding tables to a connected source.

  8. Enrich the source. This step learns what the columns mean and how the tables join. It does not run automatically. Until it finishes, questions about the data are declined. It can take up to an hour.

    CLI only
    Terminal window
    edisyl data-sources enrich <data-source-id>
    edisyl action watch <action-id>

    Ask your agent: Enrich the new data source and tell me when it’s done.

  9. Load objectives. Loading objectives starts calibration. Brainstorm them in the edisyl app, or ask your AI client to run the brainstorm tool. The tool asks which workflows the pack supports, such as a weekly review. Then it proposes questions for each step of each workflow. If you already have a list, load it with the CLI. Calibration already started in the background after steps 3 and 7.

    The assistant calls brainstorm. It adds the questions you confirm with target_questions_create.

    “Brainstorm objectives with me for the Supplier risk pack. We use it for the weekly supplier review.”

  10. Watch calibration. Pack health in the app shows each measure against its target.

    The assistant calls calibration_status.

    “How far along is calibration on the Supplier risk pack?”

Get the pack ready for people

Section titled “Get the pack ready for people”

Objectives are the questions this pack has to answer correctly, written from the work people do. Calibration tests the pack against them. For each objective the pack answers correctly, calibration stores a verified plan. Real questions then get faster and more accurate answers. See Objectives and Calibration.

  • Brainstorm first. The interview about the pack’s workflows produces much better questions than a list written without it. It runs in the app or through your AI tool. If you already have a list, load it with edisyl pack objectives structure <pack> -f questions.md.

  • Write enough of them. Start with about 10 per data source. Scale the number by the workflows and tables people use, not by the number of sources. Write about 5 per document set (a policy library, a runbook set, a spec set), and up to 10 for a large or heavily used one. Skip reference material nobody asks about.

  • Make them specific. “Which suppliers missed SLA last quarter?” is better than “Tell me about suppliers.” Claude drafts good ones from the pack’s documents. Edit them before loading.

  • Ask a few right away. Pick one to five objectives you want answered today. Ask them with edisyl ask <pack> "<question>" --wait. A weak answer shows you what to fix. See Improving answers.

  • Open the pack to users after the first calibration. Before the first calibration, answers are slower and less consistent.

  • Pause calibration during a rebuild or peak hours. Resuming needs the admin role in your organization.

    The assistant calls calibration_pause. Resuming is CLI only.

    “Pause calibration on the Supplier risk pack while we rebuild the warehouse.”

How long it takes

Section titled “How long it takes”
Stage Time
Create the pack, add documents, connect the source Minutes
Enrich a data source Up to an hour
First calibration, small pack Under an hour
First calibration, 150+ tables 4 to 6 hours. Run it overnight.

After setup

Section titled “After setup”
To Say in chat
Replace a document with a newer version “Forget the old supplier policy and teach the attached new version.” The assistant calls forget, then teach. Finding a document’s record id is CLI only.
Add or fix a definition or acronym “Define ‘OTD’ as on-time delivery.” The assistant calls lexicon_edit or lexicon_create.
Add a fact the documents lack “Teach the pack that <fact>.” The assistant calls teach.
Test the pack “Ask the pack <question>.” The assistant calls ask.

Report incorrect code

Please provide a detailed description of the incorrect code.