> ## Documentation Index
> Fetch the complete documentation index at: https://docs.driftless.icu/llms.txt
> Use this file to discover all available pages before exploring further.

# MCP: Workspace and collaboration

> The collaboration tools over MCP: leave and resolve comments on any spine object, list workspace members and workspaces, and read proposal outcome stats.

These tools let a human and an agent coordinate over the same workspace: annotate an object without editing it, find the members and workspaces a credential can reach, and see how agent proposals fared. Comment writes need the `work:write` scope; the list tools are member-level reads.

## Comments

A **comment** points AT a spine object (a topic, record, or card) and optionally at one field. It carries no governance and resolves into an edit; its text never enters the topic body. Agent-authored comments are what make the workspace collaborative between humans and agents.

| Tool                        | Actions                  | Main params                                     | Notes                        |
| --------------------------- | ------------------------ | ----------------------------------------------- | ---------------------------- |
| `driftless_context_comment` | `add`, `list`, `resolve` | `target_kind`, `target_id`, `field_key`, `body` | `open → resolved → wont_fix` |

Permissions: `add` / `resolve` need `work:write`; `list` is member-level. Availability: Available.

```text theme={"theme":"github-light"}
driftless_context_comment action:'add' target_kind:'topic' target_id:'billing-flow' field_key:'decisions' body:'This contradicts the refund invariant'
```

The killer placement is the Note to Knowledge gate: pin a precise, field-level objection at review time instead of a coarse approve or reject.

## Members and workspaces

Use these to resolve the identifiers the write tools expect: an `owner_clerk_id` for assigning a card or record, or the workspace slug a credential can target.

| Tool                   | Purpose                                        | Returns                |
| ---------------------- | ---------------------------------------------- | ---------------------- |
| `driftless_members`    | List the workspace's members                   | member id, role, email |
| `driftless_workspaces` | List the workspaces your credential can target | workspace slugs        |

Permissions: member-level reads. Availability: Available. Every tool accepts an optional `workspace` argument to target another workspace your credential is authorized for.

```text theme={"theme":"github-light"}
driftless_members   → [ { id, role, email } ]   (the valid owner_clerk_id values)
```

## Agent stats

`driftless_agent_stats` reports proposal outcomes: how many agent proposals (the Auditor's Suggested edits) merged versus were rejected, so you can read the approval rate.

| Tool                    | Purpose                | Returns                                  |
| ----------------------- | ---------------------- | ---------------------------------------- |
| `driftless_agent_stats` | Proposal outcome stats | merged vs rejected counts, approval rate |

Permission: member-level read. Availability: Available. This is the only agent-runs signal on the MCP surface; the agent-runs config and trigger surface is not reachable over OAuth/MCP.

## Permissions

| Scope         | Covers                                                 |
| ------------- | ------------------------------------------------------ |
| `work:write`  | `comment` add and resolve                              |
| member (read) | `comment` list, `members`, `workspaces`, `agent_stats` |

## Availability

| Capability                          | State                                         |
| ----------------------------------- | --------------------------------------------- |
| Comments (`add`, `list`, `resolve`) | Available                                     |
| Members and workspaces lists        | Available                                     |
| Agent proposal stats                | Available                                     |
| Manual agent-run trigger over MCP   | Not available (only `agent_stats` is exposed) |

## Related

* [Comments](/concepts/comments), [Workspaces](/concepts/workspaces) - the concepts.
* [Govern agent learning](/guides/govern-agent-learning) - comments at the review gate.
* [CLI: Context Commands](/cli/context) - comments and members from the terminal.
* [MCP & OAuth](/mcp/overview) - setup, auth, and scopes.