Developer Documentation

Get your tools talking to real facts.

MsgStack is built on the Model Context Protocol (MCP), so it exposes your organizational truth as tools any AI assistant (Claude, Cursor, ChatGPT) can call directly — no copy-pasting context into a prompt. Below is the MCP connection for developers; if you just want to ask MsgStack a question from Slack or Teams without touching config files, see the agent page (in active development).

Connect to Claude Desktop

Add MsgStack to your local AI environment in seconds. Open your Claude configuration file and inject the server endpoint.

claude_desktop_config.json

{
  "mcpServers": {
    "msgstack": {
      "url": "https://mcp.abidc.dev/mcp"
    }
  }
}

Note: For self-hosted deployments, replace the URL with your own server address (e.g., http://localhost:8001/mcp).

5-Minute Quickstart

Add the URL

Paste the MsgStack server URL into your MCP client config. No install required for using the public demo server.

Connect

Reload your MCP client (Claude Desktop, Cursor). MsgStack appears as a connected server.

Discover

Run 'list_canon_domains' to see what's already structured — Message House, Engineering Spec, Policy Shield, or your own.

Set Active

Select a domain (e.g., 'Message House') to pin your grounding context.

Ground

Use 'generate_artifact' to create cited, approved content.

Standard MCP Toolset

The MsgStack server exposes four categories of tools to your AI assistants.

Discovery & Grounding

Identify available truth layers and perform hybrid retrieval.

list_canon_domainssearch_canonget_canon_domain

Artifact Generation

Turn grounded context into marketing, legal, or technical docs.

generate_artifactbuild_ui_artifact

Governance & Review

Audit your canon layer and score third-party text for drift.

check_canon_completenessscore_canon_alignment

Session Management

Manage the stateful grounding context for the current chat.

set_active_domainreset_conversation

Glossary

Concepts & Terminology

A few terms come up constantly once you're using MsgStack. Here's the short version of each.

The Canon Layer

The organizational layer of approved truth. It sits between your raw data sources and your AI agents, acting as a governance filter for all retrieved context.

Canon Domain

A logical slice of the canon owned by a specific department or team (e.g., a Message House for Product Marketing, or a Policy Shield domain for Legal).

Grounding Type

The schema that defines a domain's shape. Message House structures brand positioning and proof points; Engineering Spec structures API contracts and SLAs; Policy Shield structures legal disclaimers and compliance language.

Canon Entry

An individual unit of truth within a domain. An entry could be a single compliance rule, a product feature claim, or an approved headline.

Derived Artifact

Any content generated by an AI agent that is grounded in the canon (e.g., a battlecard, a release note, or a press release).

Retrieval Fusion

The mechanism of combining deterministic graph traversal (finding exact matches via adjacency) with semantic vector search (finding similar context via embeddings).

Approval-gated Grounding

A governance rule that prevents AI agents from retrieving any canon entries that haven't been explicitly marked as APPROVED or LOCKED.

Governance Lifecycle

The states an entry moves through: DRAFT (private) → IN REVIEW → APPROVED (queryable) → LOCKED (verbatim required) → OUTDATED (archived).

Deeper Reference

Explore the full API documentation, contribution guidelines, and advanced deployment options on GitHub.

Full API Reference System Architecture →