Install via Your Agent

This page gives you copy-paste prompts to hand to your AI coding agent. The agent reads the prompt, installs sqry, configures its own MCP client to talk to sqry-mcp, indexes the current project, and verifies the setup by running a tool. No manual config-file editing on your part.

The prompts are idempotent — re-running them won’t break a working install. They detect the host OS, fall back gracefully when a step is already done, and print a one-line verification at the end.

Pick your agent

AgentConfig file the prompt editsUse this prompt
Claude Code~/.claude.json (or project .claude.json)Claude Code prompt
OpenAI Codex CLI~/.codex/config.tomlCodex CLI prompt
Gemini CLI~/.gemini/settings.jsonGemini CLI prompt
Cursor~/.cursor/mcp_settings.jsonCursor prompt
WindsurfSettings → MCP ServersWindsurf prompt
Any agent (auto-detect)depends on which agent runs itGeneric prompt

Claude Code

Paste this entire block into a Claude Code session:

Install sqry (semantic code search by Verivus) and configure me to use it as
an MCP client. Steps:

1. Detect the host OS. If macOS or Linux, run:
     curl -fsSL https://raw.githubusercontent.com/verivus-oss/sqry/main/scripts/install.sh \
       | bash -s -- --component all
   If Windows (PowerShell), run:
     irm https://raw.githubusercontent.com/verivus-oss/sqry/main/scripts/install.ps1 | iex
   Skip if `sqry --version` already prints a version >= 12.1.6.

2. Verify all four binaries resolve on PATH:
     sqry --version
     sqry-mcp --version
     sqry-lsp --version
     sqryd --version

3. From the project root the user is currently in, run:
     sqry index .
   If the index already exists and is current, sqry will be a no-op.

4. Configure Claude Code as an MCP client by running:
     sqry mcp setup --tool claude
   Then verify with:
     sqry mcp status

5. Append the following block to the project's CLAUDE.md (create the file
   if it doesn't exist) — only if not already present:

       ## Code Search
       Use sqry MCP tools (mcp__sqry__*) for semantic code search.
       Read sqry://docs/capability-map to find the right tool.
       Use Grep for literal text search. Use Glob for file finding.

6. Restart yourself (the user will be prompted to reload). After reload,
   verify the MCP connection by calling mcp__sqry__get_index_status and
   reporting the version + node/edge counts.

7. Optional: For long sessions or large repos, also start the daemon and
   re-configure MCP to use it:
     sqry daemon start
     sqry daemon load .
   Then edit the Claude entry in `.claude.json` so the sqry server's
   "args" list contains "--daemon".

Print a single-line summary at the end:
  "sqry vX.Y.Z installed, indexed N files, MCP wired (daemon: yes/no)."

Refuse to proceed if any step would require sudo without the user
explicitly approving it.

Codex CLI

Paste this into a Codex CLI session:

Install sqry (semantic code search by Verivus) and configure me to use it as
an MCP client. Steps:

1. Detect the host OS. If macOS or Linux:
     curl -fsSL https://raw.githubusercontent.com/verivus-oss/sqry/main/scripts/install.sh \
       | bash -s -- --component all
   If Windows (PowerShell):
     irm https://raw.githubusercontent.com/verivus-oss/sqry/main/scripts/install.ps1 | iex
   Skip if `sqry --version` already prints a version >= 12.1.6.

2. Verify: sqry --version, sqry-mcp --version, sqry-lsp --version, sqryd --version.

3. From the project root, run:
     sqry index .

4. Configure Codex as an MCP client:
     sqry mcp setup --tool codex
   This writes [mcp_servers.sqry] to ~/.codex/config.toml.
   Verify with: sqry mcp status

5. Append this block to AGENTS.md (create if missing, skip if already there):

       ## Code Search
       Use sqry MCP tools for semantic code search.
       Read sqry://docs/capability-map to find the right tool.
       Use rg for literal text search.

6. Tell the user to restart Codex so the MCP server is picked up. After
   restart, call mcp__sqry__get_index_status as a verification step.

7. Optional daemon mode: after step 6 succeeds, run
     sqry daemon start && sqry daemon load .
   then edit ~/.codex/config.toml to add `args = ["--daemon"]` under
   [mcp_servers.sqry].

Print a single-line summary at the end.
Refuse to proceed if any step would require sudo without explicit approval.

Gemini CLI

Paste this into a Gemini CLI session:

Install sqry (semantic code search by Verivus) and configure me to use it as
an MCP client. Steps:

1. Detect host OS. macOS/Linux:
     curl -fsSL https://raw.githubusercontent.com/verivus-oss/sqry/main/scripts/install.sh \
       | bash -s -- --component all
   Windows (PowerShell):
     irm https://raw.githubusercontent.com/verivus-oss/sqry/main/scripts/install.ps1 | iex
   Skip if `sqry --version` >= 12.1.6.

2. Verify: sqry --version, sqry-mcp --version, sqry-lsp --version, sqryd --version.

3. From the project root, run: sqry index .

4. Configure Gemini as an MCP client:
     sqry mcp setup --tool gemini
   This writes the sqry entry under "mcpServers" in ~/.gemini/settings.json.
   Verify with: sqry mcp status

5. Append this block to GEMINI.md (create if missing, skip if already there):

       ## Code Search
       Use sqry MCP tools for semantic code search.
       Read sqry://docs/capability-map to find the right tool.
       Use grep/rg for literal text search.

6. Tell the user to restart Gemini CLI so the MCP server is picked up.
   After restart, call mcp__sqry__get_index_status to verify.

7. Optional daemon mode: sqry daemon start && sqry daemon load .
   Then edit ~/.gemini/settings.json to set "args": ["--daemon"] for the
   sqry server.

Print a single-line summary at the end.
Refuse to proceed if any step would require sudo without explicit approval.

Cursor

Paste this into a Cursor session (Cmd+L / Ctrl+L):

Install sqry (semantic code search by Verivus) and configure Cursor to use
it as an MCP client. Steps:

1. Install via:
     curl -fsSL https://raw.githubusercontent.com/verivus-oss/sqry/main/scripts/install.sh \
       | bash -s -- --component all
   (or install.ps1 | iex on Windows). Skip if `sqry --version` >= 12.1.6.

2. Verify: sqry --version, sqry-mcp --version, sqry-lsp --version.

3. Index the current project: sqry index .

4. Edit (or create) ~/.cursor/mcp_settings.json so it contains:

       {
         "mcpServers": {
           "sqry": {
             "command": "/absolute/path/to/sqry-mcp",
             "env": {
               "SQRY_MCP_WORKSPACE_ROOT": "/absolute/path/to/this/project"
             }
           }
         }
       }

   Resolve the sqry-mcp path with `which sqry-mcp` (or `where sqry-mcp`
   on Windows). Use the actual absolute project root for
   SQRY_MCP_WORKSPACE_ROOT — `~` and relative paths are not supported.
   Preserve any other servers already in the file.

5. Tell the user to enable MCP in Cursor (Settings → Features → MCP) if
   not already on, then fully restart Cursor.

6. Print a single-line summary at the end.
Refuse to proceed if any step would require sudo without explicit approval.

Windsurf

Paste this into a Windsurf session:

Install sqry (semantic code search by Verivus) and configure Windsurf to
use it as an MCP client. Steps:

1. Install via:
     curl -fsSL https://raw.githubusercontent.com/verivus-oss/sqry/main/scripts/install.sh \
       | bash -s -- --component all
   (Windows: install.ps1 | iex). Skip if `sqry --version` >= 12.1.6.

2. Verify: sqry --version, sqry-mcp --version.

3. Index the current project: sqry index .

4. In Windsurf, open Settings → MCP Servers and add:

       {
         "sqry": {
           "command": "/absolute/path/to/sqry-mcp",
           "env": {
             "SQRY_MCP_WORKSPACE_ROOT": "/absolute/path/to/this/project"
           }
         }
       }

   Resolve the sqry-mcp path with `which sqry-mcp` (or `where sqry-mcp`).
   Use the actual absolute project root for SQRY_MCP_WORKSPACE_ROOT.

5. Tell the user to fully restart Windsurf.

6. Print a single-line summary at the end.
Refuse to proceed if any step would require sudo without explicit approval.

Generic (any agent)

If you’re not sure which agent you’re talking to, paste this. The agent will detect itself and pick the right config file:

Install sqry (semantic code search by Verivus) and configure yourself as an
MCP client. Steps:

1. Detect which agent you are. Possible values:
     claude  — you are Claude Code (writes ~/.claude.json or .claude.json)
     codex   — you are OpenAI Codex CLI (writes ~/.codex/config.toml)
     gemini  — you are Gemini CLI (writes ~/.gemini/settings.json)
     other   — print the manual config snippet for the user to apply.

2. Detect the host OS. If macOS or Linux:
     curl -fsSL https://raw.githubusercontent.com/verivus-oss/sqry/main/scripts/install.sh \
       | bash -s -- --component all
   Windows (PowerShell):
     irm https://raw.githubusercontent.com/verivus-oss/sqry/main/scripts/install.ps1 | iex
   Skip if `sqry --version` >= 12.1.6.

3. Verify: sqry --version, sqry-mcp --version, sqry-lsp --version, sqryd --version.

4. From the user's current project root: sqry index .

5. Configure yourself as an MCP client:
     sqry mcp setup --tool <claude|codex|gemini>
   Verify with: sqry mcp status.

6. If you cannot self-detect, print this manual config block:

       {
         "mcpServers": {
           "sqry": {
             "command": "/absolute/path/to/sqry-mcp",
             "env": {
               "SQRY_MCP_WORKSPACE_ROOT": "/absolute/path/to/project"
             }
           }
         }
       }

   Tell the user to add it to their agent's MCP config file.

7. Append a "## Code Search" section to whichever instructions file the
   agent reads (CLAUDE.md, AGENTS.md, GEMINI.md, or equivalent), telling
   future-you to use sqry MCP tools and to read sqry://docs/capability-map.

8. Tell the user to restart the agent. After restart, call
   mcp__sqry__get_index_status as a verification step and print the
   resulting version, node count, and edge count.

9. Optional daemon mode (run only on the user's request):
     sqry daemon start && sqry daemon load .
   Then add `--daemon` to the sqry server's args in the MCP config file.

Print a single-line summary at the end:
  "sqry vX.Y.Z installed, indexed N files, MCP wired for <agent>
   (daemon: yes/no)."

Refuse to run any step that requires sudo without explicit user approval.
Refuse to delete or overwrite unrelated entries in the MCP config file.

What the prompts do not do

Verifying after the agent finishes

Run these by hand to confirm the install:

sqry --version
sqry-mcp --version
sqry mcp status
sqry --list-languages | head

Inside your agent, ask “use sqry to show graph stats for this project”. The agent should call mcp__sqry__get_graph_stats (or equivalent) and return a structured JSON response with node and edge counts. If you see prose instead of structured tool output, the MCP connection is not active — restart the agent.

See MCP Integration for the per-agent reference, LSP Integration for editor setup, and Daemon (sqryd) for warm-graph operation.