More MCP clients

Any MCP client that speaks Streamable HTTP and can send a custom header connects to Spicrawl. Use these settings, then find your client below:

Once connected, the client lists 25 spicrawl_* tools. Set your key in the environment the client starts from, and keep it out of committed files:

export SPICRAWL_API_KEY=spicrawl_live_...   # add to your shell profile, then restart the client

To confirm the endpoint is reachable before you debug a client, run curl -i https://mcp.spicrawl.com/mcp. Without a key it answers HTTP 401 with Unauthorized: send your Spicrawl API key. Every tool is described on MCP server, and the agent skill is at https://app.spicrawl.com/skill.md (Agent skill).

Clients with their own page

• Claude Code, Cursor, Codex, OpenCode, Gemini CLI
• GitHub Copilot, Windsurf, Cline, Kilo Code
• Zed, JetBrains, Goose, Kiro
• Factory, Devin, OpenHands
• App builders, Frameworks
• Claude Desktop: see the Claude Desktop tab on MCP server. It uses the mcp-remote bridge.

Crush

Crush (Charm) supports http MCP servers. Header values go through shell expansion, so $SPICRAWL_API_KEY is read from the environment when Crush starts.

Add the server to crush.json in the project (.crush.json also works), or to the global ~/.config/crush/crush.json:

{
  "mcp": {
    "spicrawl": {
      "type": "http",
      "url": "https://mcp.spicrawl.com/mcp",
      "headers": { "Authorization": "Bearer $SPICRAWL_API_KEY" }
    }
  }
}

Crush's newer configuration format is a crushrc (./.crushrc, ./crushrc or ~/.config/crush/crushrc), a Bash file with Crush builtins. The same server there:

mcp add spicrawl --type http --url "https://mcp.spicrawl.com/mcp" \
  --header Authorization "Bearer $SPICRAWL_API_KEY"

Do not set "oauth": true on this server. Start Crush from a shell where SPICRAWL_API_KEY is set, then ask it to list its spicrawl_* tools.

Crush reads skills from .agents/skills, .crush/skills, .claude/skills and .cursor/skills in the project, and ~/.agents/skills/ and ~/.config/crush/skills/ globally. It also loads ~/.config/AGENTS.md as global instructions, and AGENTS.md is the default name of the project context file that /init writes. Install the skill with spicrawl skill install --client codex, which writes .agents/skills/spicrawl/SKILL.md.

Amp

Amp (Sourcegraph) stores MCP servers under amp.mcpServers in ~/.config/amp/settings.json, or .amp/settings.json in a project. ${VAR} is expanded in the file. amp mcp add takes no header flag, so add the entry to the settings file:

{
  "amp.mcpServers": {
    "spicrawl": {
      "url": "https://mcp.spicrawl.com/mcp",
      "headers": { "Authorization": "Bearer ${SPICRAWL_API_KEY}" }
    }
  }
}

If you add other keys to the same file, keep them in the one JSON object. Check the connection:

amp mcp doctor

It prints each server's status, including whether a workspace server still needs approval (amp mcp approve spicrawl).

Amp reads skills from ~/.config/agents/skills/, ~/.agents/skills/, ~/.config/amp/skills/, and .agents/skills/ and .claude/skills/ in the project and its parents. It reads AGENTS.md from the working directory and its parents, and from ~/.config/amp/AGENTS.md and ~/.config/AGENTS.md. spicrawl skill install --client codex writes to .agents/skills/spicrawl/SKILL.md, which Amp finds.

Kimi Code CLI

Kimi Code CLI (Moonshot AI) reads mcp.json from ~/.kimi-code/mcp.json (or $KIMI_CODE_HOME/mcp.json) for every project, and .kimi-code/mcp.json in a project. A project entry overrides a user entry of the same name. An entry with a url and no transport is an HTTP server. bearerTokenEnvVar names the environment variable that holds the key, so the file stays free of secrets:

{
  "mcpServers": {
    "spicrawl": {
      "url": "https://mcp.spicrawl.com/mcp",
      "bearerTokenEnvVar": "SPICRAWL_API_KEY"
    }
  }
}

Alternatively, "headers": { "Authorization": "Bearer spicrawl_live_..." } writes the key into the file. Servers added while a session is open join only sessions started later. Run /mcp in the session to see each server's connection status. /mcp-config adds and edits servers from the terminal UI.

Kimi Code CLI reads skills from ~/.kimi-code/skills/, ~/.agents/skills/, and .kimi-code/skills/ and .agents/skills/ in the project. It injects AGENTS.md (project root or .kimi-code/AGENTS.md, plus ~/.kimi-code/AGENTS.md and ~/.agents/AGENTS.md) into its prompt. spicrawl skill install --client codex writes .agents/skills/spicrawl/SKILL.md.

Warp

Warp takes MCP servers as JSON. A server with a url is a Streamable HTTP (or SSE) server, and headers is an object of request headers.

{
  "spicrawl": {
    "url": "https://mcp.spicrawl.com/mcp",
    "headers": { "Authorization": "Bearer spicrawl_live_..." }
  }
}

Warp reads project rules from AGENTS.md (or WARP.md) at the project root.

Qwen Code

Qwen Code has separate keys per transport. httpUrl is Streamable HTTP; url is SSE, so use httpUrl. Add the server with one command (--scope user stores it in ~/.qwen/settings.json for every project):

qwen mcp add --scope user --transport http spicrawl https://mcp.spicrawl.com/mcp \
  --header "Authorization: Bearer $SPICRAWL_API_KEY"

Your shell expands $SPICRAWL_API_KEY before Qwen Code stores it, so the key lands in the settings file. To keep it out, write the entry by hand: Qwen Code expands $VAR and ${VAR} in settings.json from your environment.

{
  "mcpServers": {
    "spicrawl": {
      "httpUrl": "https://mcp.spicrawl.com/mcp",
      "headers": { "Authorization": "Bearer ${SPICRAWL_API_KEY}" }
    }
  }
}

Project scope is .qwen/settings.json. Run qwen mcp list to check, or /mcp inside a session.

Qwen Code reads skills from ~/.qwen/skills/ and .qwen/skills/. Its context file defaults to QWEN.md; to load AGENTS.md too, set "context": { "fileName": ["QWEN.md", "AGENTS.md"] } in settings.json. Install the skill with mkdir -p .qwen/skills/spicrawl && curl -fsSL https://app.spicrawl.com/skill.md -o .qwen/skills/spicrawl/SKILL.md.

Continue

Continue configures MCP servers in YAML. Put a file in .continue/mcpServers/ at the top of the workspace, or add the same mcpServers block to config.yaml. MCP tools work in Agent mode only. For a streamable-http server, apiKey is sent as Authorization: Bearer <apiKey>:

name: Spicrawl
version: 0.0.1
schema: v1
mcpServers:
  - name: spicrawl
    type: streamable-http
    url: https://mcp.spicrawl.com/mcp
    apiKey: ${{ secrets.SPICRAWL_API_KEY }}

Or set the header yourself with requestOptions.headers:

mcpServers:
  - name: spicrawl
    type: streamable-http
    url: https://mcp.spicrawl.com/mcp
    requestOptions:
      headers:
        Authorization: Bearer ${{ secrets.SPICRAWL_API_KEY }}

${{ secrets.SPICRAWL_API_KEY }} is read from a .env file at ~/.continue/.env or at the workspace root. Add the line SPICRAWL_API_KEY=spicrawl_live_... there and keep that file out of version control. Switch to Agent mode and open the tools list to confirm the server and its spicrawl_* tools.

Augment

Augment's Auggie CLI supports remote HTTP servers with headers:

auggie mcp add spicrawl \
  --transport http \
  --url https://mcp.spicrawl.com/mcp \
  --header "Authorization:Bearer $SPICRAWL_API_KEY"

Or edit ~/.augment/settings.json:

{
  "mcpServers": {
    "spicrawl": {
      "type": "http",
      "url": "https://mcp.spicrawl.com/mcp",
      "headers": { "Authorization": "Bearer spicrawl_live_..." }
    }
  }
}

Check the result with auggie mcp list. The command form stores the expanded key in the settings file.

Auggie reads skills from ~/.augment/skills/, ~/.claude/skills/, ~/.agents/skills/ and the matching .augment/, .claude/ and .agents/ folders in the workspace. It reads AGENTS.md and CLAUDE.md as rules, including from parent directories.

Amazon Q Developer

In the IDE (VS Code, JetBrains and others), Amazon Q Developer adds HTTP servers from a form and stores them in ~/.aws/amazonq/default.json (global) or .amazonq/default.json (project).

The form writes the key into default.json; keep the project-level file out of version control. Legacy mcp.json files (~/.aws/amazonq/mcp.json and .amazonq/mcp.json) are still read while useLegacyMcpJson is true, its default. If Amazon Q asks you to authorize in a browser, the server was treated as OAuth: check the header is set.

Google Antigravity

Google Antigravity uses serverUrl, not url, for remote servers (it also covers Streamable HTTP). Edit mcp_config.json:

• Global: ~/.gemini/config/mcp_config.json
• Workspace: .agents/mcp_config.json
• Antigravity 2.0 app: Settings → Customizations → Open MCP Config
• IDE: in the Agent panel, Additional options → MCP Servers → Manage MCP Servers → View raw config

{
  "mcpServers": {
    "spicrawl": {
      "serverUrl": "https://mcp.spicrawl.com/mcp",
      "headers": { "Authorization": "Bearer spicrawl_live_..." }
    }
  }
}

The file holds the key in plain text, so keep the workspace copy out of version control. Restart the assistant or reload MCP servers, then ask it to call spicrawl_usage_summary or any spicrawl_* tool to confirm the connection. Antigravity has its own Agent Skills feature; see its documentation for the skill folder.

Troubleshooting

More causes and fixes are in the Troubleshooting table on MCP server.