About ravendb-mcp
A local, read-only MCP server for developers and agents diagnosing RavenDB clusters. Its 21 tools inspect cluster, database, index, task, storage, performance, log, and data state, with paged results and progressive detail.
Key features
- 0121 read-only diagnostic tools
- 02Inspect clusters, databases, indexes, and tasks
- 03Query performance, storage, logs, and data
- 04Serve RQL documentation as resources
- 05Paged results and progressive detail
Use cases
- 01Check RavenDB cluster and database status
- 02Diagnose index, task, or performance issues
- 03Inspect query results and document data as needed
- 04Export logs or debug packages for local review
MCP support
The local MCP server exposes 21 read-only tools to clients and provides version-aware RQL documentation as rql:// resources.
Configuration
Local server config is the same for every client. Copy it into your client’s MCP config file.
We haven’t collected a package or remote endpoint for this server. Follow the instructions in the project README.
Tools
Local servers expose tools only once running on your machine, and the maintainer has not provided a tool list.
Details
From the project READMERavenDB MCP
A local, read-only MCP diagnostics server for RavenDB. Point it at a cluster and your AI agent can inspect cluster, database, index, task, storage, and performance state, plus logs, support packages, and your data, across 21 read-only tools.
Quick start
Launch it with npx (Node.js), dnx (.NET SDK), or a prebuilt binary from Releases, then add it to your client. Set RAVENDB_URLS to your cluster; the examples use http://localhost:8080.
Claude Code
claude mcp add ravendb --scope user --env RAVENDB_URLS=http://localhost:8080 -- npx -y @ravendb/mcp
Or install the plugin, which prompts for your cluster URL:
/plugin marketplace add ravendb/ravendb-mcp
/plugin install ravendb-mcp@ravendb
Claude Desktop
Download ravendb-mcp-<version>.mcpb from Releases, then install it under Settings > Extensions > Advanced settings > "Install Extension…" (Anthropic's guide). It prompts for your cluster URL, so there is no config to edit and no Node or .NET needed. Config-file setup is in INSTALL.md.
Codex
Add to ~/.codex/config.toml:
[mcp_servers.ravendb]
command = "npx"
args = ["-y", "@ravendb/mcp"]
env = { RAVENDB_URLS = "http://localhost:8080" }
Cursor
Click the button (installs with http://localhost:8080; edit for another cluster), or add .cursor/mcp.json:
{
"mcpServers": {
"ravendb": {
"command": "npx",
"args": ["-y", "@ravendb/mcp"],
"env": { "RAVENDB_URLS": "http://localhost:8080" }
}
}
}
Then ask the agent “list my RavenDB databases.” Any of these can use dnx RavenDB.Mcp --yes or a binary path in place of npx -y @ravendb/mcp. Secured (HTTPS) clusters, other OSes, and the full configuration reference are in INSTALL.md.
Features
Developer experience
- One-line install with
npx,dnx, or a prebuilt binary. - 21 tools, one per subject area, each with parameters to select what you need (reduced from 74).
- Version-aware RQL docs are served as
rql://resources; a failed query returns the parser error and where to look.
Safety
- Read-only: no write, patch, or delete tools, and mutating queries are rejected.
- Connection-string secrets (passwords, cloud keys, SAS tokens, AI-provider keys, certificates) are masked in the returned JSON.
- Access is scoped to the client certificate's RavenDB permissions.
get_document_dataandrun_queryreturn real cluster data into the agent's context — you choose which cluster to expose. See SECURITY.md.
Context-window friendly
- Progressive disclosure: tools return an overview by default and drill into detail on request.
- Results are paged:
run_querytakesstartandpageSize(up to 128 rows), and the list tools takepageSize. - Log exports and debug packages are written to an artifact file on disk; the tool returns its path and byte size, not the contents.
- Query results omit per-document metadata unless you request it.
Runs anywhere
- Distributed via npm, NuGet, GitHub Releases, and the MCP Registry.
- Self-contained binaries for Windows, macOS, and Linux (x64 and arm64), with no runtime to install.
- Run a separate instance per cluster for multi-cluster setups.
Configuration
One server instance targets one cluster. The only required setting is the URL; pass it (and the optional certificate / artifacts settings) as environment variables or a --config JSON file:
| Variable | Required | Purpose |
|---|---|---|
RAVENDB_URLS | yes | Cluster node URL(s), comma-separated, the client fails over across them. |
RAVENDB_CERTIFICATE_PATH / _PASSWORD | no | Client .pfx for secured (HTTPS) clusters. Omit for unsecured. |
RAVENDB_ARTIFACTS_PATH | no | Where exported files are written (defaults to a temp folder). |
Full configuration reference, including secured clusters and multi-cluster setups, is in INSTALL.md.
Tools
21 read-only tools (snake_case). Most are facet tools: one per subject that takes selectors and returns only the sections you ask for.
- Facets:
get_cluster_overview,get_server_config,get_server_resources,get_network_details,get_database_stats,get_database_config,get_index,get_tasks,get_live_workload,inspect_storage,get_document_data,sample_live_feed,wait_for_completion,collect_debug_package,get_ai_agents - Singletons:
list_databases,get_database_record,get_notifications,run_query,list_compare_exchange,export_server_logs
Large or binary output (logs, debug packages) is written to a file and returned as a reference.
Privacy
RavenDB MCP runs locally and sends no data to RavenDB or Anthropic: no telemetry, no analytics. It connects only to the cluster URL(s) you configure (and to npm/NuGet when you install via npx/dnx), returns results to your local agent, masks connection-string secrets, and writes exports only to a local folder you control. Details in PRIVACY.md; RavenDB's company policy is at ravendb.net.
License
MIT. See LICENSE. Building from source is covered in INSTALL.md.
mcp-name: io.github.ravendb/ravendb-mcp
Comments
New to MCP?
From what the protocol is to installing your first server, these answers cover most first questions.
An open protocol that defines how AI applications connect to external tools and data in a uniform way. Clients talk to the model; MCP servers provide tools, resources and prompt templates. They exchange standard messages, so one server can be reused by many clients.