Your agent, not ours
Claude Code, claude.ai, Claude Desktop, Cursor or any MCP client. Corpole provides the context; the agent you already use does the work.
Governed context
The agent cannot be steered by a decision you already retired.
When the agent asks Corpole what applies to its task, it gets the Authority records, their acceptance criteria, the constraints and accepted decisions around them, and the standards of the tenant. Retired records are not in the answer. That is the guarantee, and it is the reason the rest of the system exists.
- Works with the agent you already use
- Only Authority records bind
- Every read and proposal recorded
GOVERNED CONTEXT · MCP
get_work_contextAuthority onlycriteria + constraints + decisions
Requirements
Decisions
Standards
Tests
CODING AGENTClaude Code, Cursor…within the agent policy
How the agent works with it
Smaller context, and a correct one.
The two usually pull against each other. Here they do not, because the records that are left out are the ones that stopped applying.
Scoped to the task
Asked about a story, Corpole returns its requirements with their chain up to the objective, the criteria to meet, constraints, dependencies, conflicts, tests and recent changes.
Standards first
The agent reads the standards that apply — tenant-wide and project — at the start of a session, minus any a project has waived by an accepted decision.
Proposals, not writes
An agent can propose a requirement, a decision, a relation or a test, ask a question or report a defect or risk. It cannot make anything binding.
Results come back
The agent reports test results and acceptance results, links its pull requests and commits, and moves its story to code review within the flow rules.
Everything is on the record
Agent activity lists every read and proposal with the connection it came through; history marks each change the agent made.
What the agent may do alone
An autonomy policy per project, enforced before the action.
Some changes are routine for an agent; some should wait for a person. You decide which, per project. The agent reads the policy through MCP, Corpole refuses tool calls it forbids, and the repository kit turns it into Claude Code permission rules.
Allowed, ask first, forbidden
Writing migrations, adding dependencies, changing CI/CD, security code, tests or public APIs, merging pull requests — each action gets its own level.
Governed code areas
Onboarding maps code modules to the records that govern them. An edit in an ungoverned area is questioned, or blocked when the project enforces it.
Never beyond the person
A connection carries at most the permissions of the person who created it. Revoke it on the Agents page and the next call is refused.
What this does not do
Three things worth being plain about.
It does not decide what is correct
The system surfaces conflicts and tells you what currently binds. Which record should win is a judgement, and it stays with the people accountable for it.
It does not let an agent promote itself
Whatever an agent proposes — a requirement, a decision, a relation — arrives as a Draft or a proposal. Nothing binds until a person approves it.
It does not act beyond your permissions
A connection acts for the person who created it and carries at most that person's permissions in the project. A read-only connection cannot propose anything.
Connect your coding agent
One MCP server, three ways in.
Claude Code, claude.ai, Claude Desktop, Cursor and any MCP client read the same Authority records. Sign in with OAuth, pick a project, done.
Remote MCP server
Streamable HTTP at corpole.com/api/mcp, with OAuth or a project token. Listed in the official MCP Registry as com.corpole/requirements.
Connection guideClaude Code plugin
/corpole:init, /corpole:context, /corpole:check and /corpole:onboard, plus a hook that checks each edit against the governed code areas.
Plugin and repository kitGitHub check
Every pull request has to cite existing, non-retired requirements. A retired decision cannot come back through a merge.
How the check worksThe guarantee, in one line
A decision you retired last month cannot reach the run you start today.
That is the whole claim. Everything else in the product exists so that this one holds.