Corunner Engineering Memory MCP

Connect AI coding agents to Corunner’s permission-aware engineering memory.

Corunner Engineering Memory MCP gives compatible AI coding agents the organizational context they cannot get from a repository alone. Use it to retrieve architecture decisions, requirements, repository history, related blockers, and change-validation findings while you work.

This is different from the MCP Servers under Integrations. Those connections let Corunner use third-party MCP servers. Corunner Engineering Memory MCP lets Codex, Claude Code, Cursor, and other agents query Corunner.

Server endpoint

Use the hosted endpoint below unless your organization provides a different Corunner deployment:

https://api.corunner.ai/mcp

What the server provides

All current tools are read-only. corunner.check_change analyzes a proposed change but does not modify your repository.

Tool	Use it to
corunner.get_context	Retrieve relevant decisions, requirements, history, incidents, and code targets before substantial work.
corunner.get_decisions	Check active, historical, superseded, or rejected architecture decisions.
corunner.get_requirements	Recover requirements and acceptance criteria from connected work items.
corunner.explain_code_history	Understand related change requests and check runs behind unfamiliar code.
corunner.get_related_incidents	Find unresolved blockers and relevant failed checks.
corunner.check_change	Validate a proposed diff against supported organizational evidence before finalizing.

Recommended workflow

1. Start substantial work with corunner.get_context. Include the canonical repository remote, task, and likely files.
2. Use corunner.get_decisions when architecture choices or historical alternatives matter.
3. Use corunner.get_requirements when the task includes a work-item ID.
4. Use corunner.explain_code_history or corunner.get_related_incidents for unfamiliar or failure-prone code.
5. Before finalizing, call corunner.check_change with the actual diff or a concrete change summary.
6. Treat partial, authorization_limited, repository_unresolved, and insufficient_context as limitations, not successful validation.

Authentication and access

Corunner uses OAuth 2.1 authorization code flow with PKCE. When an agent connects, sign in, choose the Corunner workspace, and approve only the scopes the agent needs.

Access is limited by the intersection of your OAuth scopes, workspace membership, role and resource policy, connected-provider access, repository access, and source visibility. The MCP server does not grant access you do not already have in Corunner.

Never put bearer tokens in a repository file. Keep authorization grants user-specific and let the agent manage the OAuth flow.

Understand results

• ok: relevant evidence was returned without a known source failure.
• partial: useful evidence was returned, but one or more sources were unavailable or filtered.
• no_evidence: no relevant supported evidence was found; this does not prove that no decision exists.
• repository_unresolved: Corunner could not safely map the repository. Send its canonical remote.
• authorization_limited: permissions excluded potential evidence.
• insufficient_context: provide more files, symbols, a diff, or a concrete change summary.

no_known_conflict means the supplied facts did not trigger a supported conflict. It is not a correctness, security, or policy certification.

Connect an agent

Codex

Connect Codex CLI and the Codex IDE extension.

Claude Code

Add Corunner to Claude Code with HTTP transport.

Cursor

Configure Corunner in Cursor’s MCP settings.

Related

See MCP Servers to connect a third-party MCP server to Corunner.