seq-mcp

A minimal, token-efficient MCP server for querying a Seq structured-log server. It exposes Seq SQL querying (including group by aggregates) plus two convenience tools, and returns terse plain text rather than raw JSON to keep token usage low.

Tools

Tool	Description
seq_query(sql, from?, to?, signal?)	Run a Seq SQL query. Supports aggregates/grouping, e.g. select count(*) as Count from stream group by @Level. from/to are optional ISO-8601 UTC bounds (default: last 24h). Optionally scope to a saved signal id. Results capped at 100 rows.
recent_errors(minutes?)	Most recent Error/Fatal events (up to 20) as Id · Time · Level · Message lines, newest first (default: last 30 min).
search_events(filter, count?, signal?)	Events matching a Seq filter expression (e.g. @Exception like '%timeout%'), newest first (default: 30, last 24h). Optionally scope to a saved signal id.
get_event(id)	Full detail for one event (the Id from the list tools): rendered message, exception/stack trace, and all properties.
list_signals(nameFilter?)	Saved signals as Id · Title lines; the Id can scope seq_query/search_events. Pass a name substring to narrow the list.

Install

dotnet tool install -g SeqMcp

This installs the seq-mcp command, which runs in one of three modes:

• CLI (seq-mcp <command>, the default with no flags) — run a single query from a shell, script, or agent and print plain text. See CLI below.
• stdio (seq-mcp --stdio) — the MCP client launches and manages the process, speaking JSON-RPC over stdin/stdout. Recommended for local single-user clients like Claude Code.
• HTTP (seq-mcp --http) — a long-running Streamable HTTP server on http://localhost:5250 (override with ASPNETCORE_URLS) that clients connect to by URL. Use for Claude Desktop and other clients that connect by URL.

CLI

The same capabilities are available as shell commands over the same core code, so behaviour and defaults match the MCP tools exactly. Run seq-mcp --help to list them, or seq-mcp commands for a machine-readable JSON description of every command, argument, and option (the CLI analogue of MCP's tools/list — point an agent at this to discover the surface in one call).

seq-mcp query "select count(*) as Count from stream group by @Level"
seq-mcp errors --minutes 30
seq-mcp search "@Exception like '%timeout%'" --count 30
seq-mcp event event-abc123
seq-mcp signals --filter checkout

Output goes to stdout; errors go to stderr with a non-zero exit code. Configure the server with the same SEQ__SERVERURL / SEQ__APIKEY environment variables (see Configuration).

Configuration

Setting	Environment variable	Default
Seq server URL	SEQ__SERVERURL	http://localhost:5341
Seq API key	SEQ__APIKEY	(none)

The API key is sent as the X-Seq-ApiKey header. It is optional if your Seq instance allows anonymous read; when the key is missing or Seq is unreachable, tool calls return a graceful error message rather than failing the server.

Use with Claude Code

stdio (recommended) — Claude launches and manages the process:

claude mcp add seq -e SEQ__SERVERURL=https://seq.example.com -e SEQ__APIKEY=your-api-key -- seq-mcp --stdio

HTTP — run the server yourself, then point Claude at the URL:

SEQ__SERVERURL=https://seq.example.com SEQ__APIKEY=your-api-key seq-mcp --http   # in one terminal
claude mcp add --transport http seq http://localhost:5250

Example MCP client config (HTTP)

{
  "mcpServers": {
    "seq": {
      "type": "http",
      "url": "http://localhost:5250"
    }
  }
}

Claude Desktop extension (.mcpb)

The server is also packaged as a one-click MCPB desktop extension. The bundle ships a self-contained Windows executable, so the target machine needs neither the .NET runtime nor seq-mcp on its PATH.

To install it in Claude Desktop:

1. Download seq-mcp.mcpb from the latest release.
2. Open Settings → Extensions → Advanced settings → Install Extension.
3. Select the downloaded seq-mcp.mcpb.
4. When prompted, enter your Seq server URL and (optionally) Seq API key. These are stored by Claude Desktop and passed to the server as the SEQ__SERVERURL / SEQ__APIKEY environment variables, so no secrets are baked into the bundle.

Building the bundle yourself

Requires the .NET SDK and Node.js:

./build-mcpb.ps1

This publishes the self-contained executable and writes dist/seq-mcp.mcpb. Pushing a v*.*.* tag runs the same build in CI and attaches the .mcpb to a GitHub Release.

Build from source

dotnet build seq-mcp.sln -c Release
dotnet run --project src/SeqMcp -- --help          # CLI (default)
dotnet run --project src/SeqMcp -- --stdio         # MCP over stdio
dotnet run --project src/SeqMcp -- --http          # MCP over HTTP

License

MIT