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

# MCP server

> Give AI tools like Claude Code, Cursor, and Cline direct access to your org's docs and Salesforce intelligence — reads, reviewed writes, and guide recording over MCP.

Every LivingContext organization exposes a **Model Context Protocol (MCP) server**. Point your AI coding tool at it and that tool can search your docs, query your Salesforce catalog, draft doc changes, and even drive the browser capture extension — all under your organization's permissions.

```text theme={null}
https://<your-org>.livingcontext.ai/api/mcp
```

The endpoint speaks **streamable HTTP JSON-RPC**, the standard MCP transport supported by modern clients.

## Create a key

<Steps>
  <Step title="Open Settings → MCP server">
    Keys are managed per organization. Name the key (e.g. "Claude Code"), optionally give it an expiry, and click **Create key**.
  </Step>

  <Step title="Copy the key">
    Keys are prefixed `lcmcp_` and **shown once** — copy it immediately and store it like any credential.
  </Step>

  <Step title="Paste the config snippet">
    The page provides a ready-to-paste `mcpServers` snippet — with your new key already substituted — that works in `.mcp.json` and the MCP settings of Claude Code, Cursor, and Cline alike, plus an optional rules snippet for `CLAUDE.md` / Cursor Rules so your agent reaches for LivingContext automatically.
  </Step>
</Steps>

For example, with Claude Code:

```bash theme={null}
claude mcp add --transport http livingcontext \
  https://your-org.livingcontext.ai/api/mcp \
  --header "Authorization: Bearer lcmcp_your_key_here"
```

## Security model

* Keys are **org-scoped** — a key only reaches the organization it was created in.
* Your **membership is re-checked on every call**; if you leave the org, the key stops working.
* **Writes respect your role** — the MCP server can't do anything you couldn't do in the app. See [Roles and permissions](/admin/roles-and-permissions).
* Write tools never publish directly: they land as [change requests](/guides/editing-and-change-requests) or pull requests for human review.

## Tools

The server exposes 20+ tools across five areas.

### Reads

| Tool                   | What it does                                      |
| ---------------------- | ------------------------------------------------- |
| `resolve_doc_id`       | Find a doc's identifier from a path or title      |
| `get_doc_content`      | Read a doc's full content                         |
| `get_metadata_context` | Grounding context for a Salesforce component      |
| `list_recent_changes`  | Recent documentation changes                      |
| `suggest_metadata`     | Suggest Salesforce components relevant to a topic |

### Writes

| Tool                 | What it does                                                                                                                 |
| -------------------- | ---------------------------------------------------------------------------------------------------------------------------- |
| `create_doc`         | Draft a new doc — lands as a change request                                                                                  |
| `update_doc`         | Edit an existing doc — lands as a change request                                                                             |
| `set_doc_metadata`   | Tag a doc to Salesforce components (applies immediately)                                                                     |
| `submit_for_review`  | Ship the change request — opens a PR                                                                                         |
| `upload_guide_asset` | Upload an asset for a guide                                                                                                  |
| `create_guide`       | Publish a step-by-step guide from agent-recorded steps — same pipeline as the capture extension, landing as a change request |

### Guide recording

Drive the [browser capture extension](/capture/overview) remotely:

| Tool                         | What it does                                     |
| ---------------------------- | ------------------------------------------------ |
| `start_guide_recording`      | Start a capture session in the connected browser |
| `stop_guide_recording`       | Stop the session                                 |
| `get_guide_recording_status` | Poll recording state                             |

### Org intelligence

The same Salesforce analysis available to the [workspace agent](/agents/workspace-agent):

| Tool                           | What it does                           |
| ------------------------------ | -------------------------------------- |
| `find_metadata_usages`         | Where a component is referenced        |
| `list_metadata_dependencies`   | What a component depends on            |
| `search_org_source`            | Full-text search across org source     |
| `list_recent_metadata_changes` | Recently changed components            |
| `get_metadata_change_diffs`    | Before/after diffs                     |
| `get_object_schema`            | Object fields, types, relationships    |
| `find_unreferenced_fields`     | Fields nothing references              |
| `who_can_access`               | Who can see or edit an object or field |
| `get_org_debt_signals`         | Technical-debt indicators for the org  |

### Connections

Check what's wired up before reading or writing (connecting itself happens in the app UI):

| Tool               | What it does                                                                    |
| ------------------ | ------------------------------------------------------------------------------- |
| `list_connections` | Connection status for each integration (GitHub, GitLab, Confluence, Salesforce) |
| `list_docs_repos`  | The git repositories writes may target                                          |

## Example: docs maintenance from your editor

1. Add the MCP server to Claude Code with your `lcmcp_` key.
2. Ask: *"Our discount approval flow changed — update the docs that describe it."*
3. The tool searches your docs, reads the current pages, pulls fresh metadata context, and drafts edits via `update_doc`.
4. The edits arrive in LivingContext as **change requests** — you review and ship them like any other change.

<Note>
  Nothing bypasses review. Even a fully automated MCP client ends at the same gate as every agent: a human approves before anything merges.
</Note>

<Warning>
  Treat `lcmcp_` keys like passwords. They're shown once at creation — if a key leaks, revoke it in **Settings → MCP** and create a new one.
</Warning>

<Note>
  Prefer the terminal? The [`lc` command-line client](/integrations/cli) wraps these same tools — docs search, reviewed writes, and Salesforce intelligence — using your `lcmcp_` key.
</Note>