Corpole documentation

Overview

Corpole is a requirements authority for AI-assisted software delivery. Teams capture objectives, requirements and constraints per project and govern their state; coding agents read the binding ones through a remote MCP server at https://corpole.com/api/mcp (Streamable HTTP) or a REST endpoint.

Concepts

Tenants and projects

Records live inside a project, projects inside a tenant. Every agent connection is limited to exactly one project. People join a tenant by invitation (Members → Invite someone) with one of four roles: owner, admin, member (edit records) or viewer (read only).

Record types

• Objective OBJ-… (Intent)
• Business Requirement BR-… (Requirements)
• Stakeholder Requirement SR-… (Requirements)
• Functional Requirement FR-… (Requirements)
• Non-functional Requirement NFR-… (Requirements)
• Technical Requirement TR-… (Requirements)
• Project Requirement PR-… (Requirements)
• Product Constraint PC-… (Constraints)
• Business Constraint BC-… (Constraints)
• Project Constraint PJC-… (Constraints)

Authority states

Agents treat only Authority records as binding, never implement Retired ones and surface conflicts instead of choosing a winner.

Connect your AI tool

claude.ai and Claude Desktop

1. Open Settings → Connectors → Add custom connector.
2. Enter the URL https://corpole.com/api/mcp.
3. Sign in to Corpole, choose the project and the permissions, and approve.

Claude Code plugin (recommended for Claude Code)

Adds the MCP tools, loads the Authority records at session start, and provides /corpole:context, /corpole:check and /corpole:init.

# 1. Create a project token: Corpole → project → AI Context → Generate token
export CORPOLE_TOKEN="crp_..."

# 2. In Claude Code
/plugin marketplace add Global-Agent-Factory/compedans-platform
/plugin install corpole@corpole

Claude Code, MCP only (OAuth sign-in)

claude mcp add --transport http corpole https://corpole.com/api/mcp
# then run /mcp, select corpole and choose Authenticate

Cursor, VS Code and other MCP clients

Add a remote (Streamable HTTP) server. Clients that support OAuth need only the URL; otherwise send a project token:

{
  "mcpServers": {
    "corpole": {
      "url": "https://corpole.com/api/mcp",
      "headers": {
        "Authorization": "Bearer ${CORPOLE_TOKEN}"
      }
    }
  }
}

REST and CLAUDE.md fallback

curl -H "Authorization: Bearer $CORPOLE_TOKEN" \
  "https://corpole.com/api/context?format=markdown"      # Authority records as Markdown
# JSON: omit format; all states: add mode=all

MCP tools

Read tools carry readOnlyHint: true; propose_record and report_acceptance_result carry readOnlyHint: false and destructiveHint: false (they only add Drafts or record verification results). The server also sends instructions: fetch Authority context before coding, ignore Retired records, surface conflicts, cite record codes.

Permissions

OAuth connections get their permissions on the consent screen (propose requires a non-viewer role). Project tokens get read, plus propose if the box is ticked when the token is generated.

OAuth details

Repository kit and GitHub check

Claude Code has no built-in precedence between sources; it follows its instructions. The repository kit sets them: a CLAUDE.md section that ranks Corpole Authority records above documents in the repository, .claude/settings.json that offers the plugin to the team, and optionally a PR template and a GitHub Action. Run /corpole:init in the repository or copy the files from AI Context → Repository kit.

The GitHub check fails a pull request that cites no record code (unless it has the label no-requirement), cites an unknown code or a Retired record, and warns on Draft or Approved ones. It needs the repository secret CORPOLE_TOKEN.

Security and data

• Every connection is limited to one project; read-only unless the propose permission is granted.
• Tokens are stored only as SHA-256 hashes. Project tokens can be rotated or revoked; OAuth connections can be disconnected at any time.
• Agents cannot change or delete records. Proposals are Drafts that a person must review.
• Every agent read and proposal is recorded in the project's activity log (agent, time, action, record codes).
• Data is stored in PostgreSQL in the EU (Frankfurt); the application runs in Frankfurt.
• Optional AI assistance in the project setup wizard sends the project brief and uploaded documents to Anthropic's API to suggest or extract records; nothing is saved without the user's confirmation.

See the privacy policy and terms.

Troubleshooting