Valara

Connect Claude, Cursor, and other AI agents to Valara's review engine over the Model Context Protocol.

Valara hosts a remote Model Context Protocol server so AI agents (Claude, Cursor, VS Code, Codex, and any MCP client) can run and read appraisal reviews as tools, without you writing a line of integration code. It is the same review engine behind the REST API, exposed as agent tools.

https://getvalara.com/api/mcp

The transport is MCP Streamable HTTP. Authentication uses the same first-party API keys as the REST API, passed as a bearer token, so a single vlr_live_… key works for both. See Authentication to create one.

Connect a client

The server is hosted, so there is nothing to install: each client just needs the URL and your API key. Create a key under Settings → API Keys with the scopes you want the agent to have (reviews:read for read-only triage, add reviews:write to let it submit), then drop it into the client's config.

Claude Desktop / Cursor / generic clients

Most clients accept a mcpServers block. Use the mcp-remote bridge, which forwards your key as an Authorization header:

claude_desktop_config.json

{
  "mcpServers": {
    "valara": {
      "command": "npx",
      "args": [
        "-y",
        "mcp-remote",
        "https://getvalara.com/api/mcp",
        "--header",
        "Authorization: Bearer vlr_live_your_key_here"
      ]
    }
  }
}

Cursor (~/.cursor/mcp.json) can also connect to the URL directly with headers:

~/.cursor/mcp.json

{
  "mcpServers": {
    "valara": {
      "url": "https://getvalara.com/api/mcp",
      "headers": { "Authorization": "Bearer vlr_live_your_key_here" }
    }
  }
}

Claude Code (CLI)

claude mcp add --transport http valara https://getvalara.com/api/mcp \
  --header "Authorization: Bearer vlr_live_your_key_here"

Replace vlr_live_your_key_here with the secret shown once when you created the key. Restart the client and the Valara tools appear.

Tools

Tool	Scope	What it does
`submit_review`	`reviews:write`	Start a review from a PDF URL. Returns a review id immediately.
`get_review`	`reviews:read`	Check a review's status (`processing` → `completed`).
`get_review_result`	`reviews:read`	Fetch the full scored result once completed.
`get_findings`	`reviews:read`	Flat, severity-tagged list of issues with citations: the fastest triage.
`list_reviews`	`reviews:read`	List your reviews, newest first (cursor paginated).
`delete_review`	`reviews:delete`	Delete your ownership of a review.

Each tool scopes to the API key's owner and enforces the scopes above, exactly like the REST API. A tool call that needs a scope the key lacks returns an error the agent can read, rather than doing any work.

The async contract

Reviews take minutes, so submit_review does not block. The typical agent flow is:

submit_review with a blob_url and review_type → returns a review id.
Poll get_review until status is completed.
get_findings (for triage) or get_review_result (for the full DTO).

Re-submitting the same PDF is a no-op that returns the existing review (reviews are content-addressed by the PDF's hash), so an agent can safely retry. Each new review deducts credits from the key owner's balance.

Notes

Key management is not exposed over MCP. Creating or revoking keys requires a dashboard or CLI session, never an API key, so a leaked key can never mint new ones. Manage keys at Settings → API Keys.
Review tool calls before approving them. As with any MCP server, prefer a client that confirms tool calls, and be cautious when combining Valara with other servers, to avoid prompt-injection from untrusted content.
Prefer the stream. Agents that want live progress can also use the REST event stream; MCP tools follow the poll model above.