MCP Server

CodeWall exposes a Model Context Protocol (MCP) server so AI agents — Claude Desktop, Claude Code, IDE assistants, or anything that speaks MCP — can read your organisation's pentest data and launch single-finding tests, using a scoped API key instead of your login.

Everything is scoped to the organisation that owns the API key and PII-redacted according to your org's privacy policy. Read scopes are strictly read-only; two optional action scopes let an agent launch a retest or red-team agent on an existing finding (never a full pentest).

Endpoint

https://api.codewall.ai/mcp

The server uses the Streamable HTTP transport (stateless, JSON responses).

1. Create an API key

API keys are managed by org owners and admins.

1. Go to Settings > API Keys
2. Click Add > API Key
3. Give it a label (e.g. "Claude Desktop")
4. Select the scopes to grant (read scopes are pre-selected; action scopes are opt-in)
5. Optionally set an expiry date
6. Click Create Key
7. Copy the key immediately — the cwk_… token is shown only once

The dialog also shows a ready-to-paste MCP client configuration with the key filled in.

2. Connect your client

CodeWall is a standard remote (Streamable HTTP) MCP server, so any MCP-capable client can connect — pass the endpoint and your key as a Bearer header. Examples for the most common clients follow; if yours isn't listed, use the generic configuration or the stdio bridge.

Replace cwk_your_key_here with the key you created above throughout.

Claude Code

claude mcp add --transport http codewall https://api.codewall.ai/mcp \
  --header "Authorization: Bearer cwk_your_key_here"

Claude Desktop

Add to your claude_desktop_config.json:

{
  "mcpServers": {
    "codewall": {
      "url": "https://api.codewall.ai/mcp",
      "headers": { "Authorization": "Bearer cwk_your_key_here" }
    }
  }
}

Codex CLI

Codex connects to remote servers through the mcp-remote bridge. Add to ~/.codex/config.toml:

[mcp_servers.codewall]
command = "npx"
args = [
  "-y", "mcp-remote", "https://api.codewall.ai/mcp",
  "--header", "Authorization: Bearer cwk_your_key_here",
]

Cursor

Add to ~/.cursor/mcp.json (global) or .cursor/mcp.json (per-project):

{
  "mcpServers": {
    "codewall": {
      "url": "https://api.codewall.ai/mcp",
      "headers": { "Authorization": "Bearer cwk_your_key_here" }
    }
  }
}

VS Code (Copilot agent mode)

Add to .vscode/mcp.json:

{
  "servers": {
    "codewall": {
      "type": "http",
      "url": "https://api.codewall.ai/mcp",
      "headers": { "Authorization": "Bearer cwk_your_key_here" }
    }
  }
}

Cline

In Cline's MCP Servers > Configure (cline_mcp_settings.json):

{
  "mcpServers": {
    "codewall": {
      "type": "streamableHttp",
      "url": "https://api.codewall.ai/mcp",
      "headers": { "Authorization": "Bearer cwk_your_key_here" }
    }
  }
}

Goose

Add a remote (Streaming HTTP) extension to ~/.config/goose/config.yaml:

extensions:
  codewall:
    type: streamable_http
    uri: https://api.codewall.ai/mcp
    headers:
      Authorization: Bearer cwk_your_key_here

Generic / any MCP client

Most clients accept a URL plus headers under an mcpServers map:

{
  "mcpServers": {
    "codewall": {
      "url": "https://api.codewall.ai/mcp",
      "headers": { "Authorization": "Bearer cwk_your_key_here" }
    }
  }
}

Clients without remote support

If a client only supports local (stdio) servers, bridge to the remote endpoint with mcp-remote:

{
  "mcpServers": {
    "codewall": {
      "command": "npx",
      "args": [
        "-y", "mcp-remote", "https://api.codewall.ai/mcp",
        "--header", "Authorization: Bearer cwk_your_key_here"
      ]
    }
  }
}

That's it — the agent can now call the tools below.

Scopes

A key only exposes the tools its scopes allow. Each tool checks its scope on every call.

Read scopes

Action scopes

These let a key start a test on an existing finding. They cannot launch a full pentest. Off by default.

Tools

Result limits are capped server-side (findings 5000, runs/tests 500) regardless of the limit you pass.

Typical workflows

Triage the latest findings

"Show me the open critical findings from CodeWall."

The agent calls list_findings(severity="critical", status="open").

Drill into a run

"What did run run-20260602-132211-9e7d15af find?"

The agent calls get_run(...) then list_findings(run_id="run-20260602-132211-9e7d15af").

Re-verify a fix

"Retest finding #213."

The agent calls run_retest(finding_id=2329). When the retest can't reproduce the issue, the finding is automatically marked fixed. Track progress with list_tests or get_run.

Calling the server directly

The server is plain JSON-RPC over HTTP, so you can also call it with curl. Initialise, then call a tool:

# initialize
curl -s https://api.codewall.ai/mcp/ \
  -H "Authorization: Bearer cwk_your_key_here" \
  -H "Content-Type: application/json" \
  -H "Accept: application/json, text/event-stream" \
  -d '{"jsonrpc":"2.0","id":1,"method":"initialize",
       "params":{"protocolVersion":"2025-06-18","capabilities":{},
                 "clientInfo":{"name":"curl","version":"0"}}}'

# call a tool
curl -s https://api.codewall.ai/mcp/ \
  -H "Authorization: Bearer cwk_your_key_here" \
  -H "Content-Type: application/json" \
  -H "Accept: application/json, text/event-stream" \
  -d '{"jsonrpc":"2.0","id":2,"method":"tools/call",
       "params":{"name":"list_findings","arguments":{"severity":"critical","limit":5}}}'

Security

• Keys are stored as SHA-256 hashes — CodeWall never stores the plaintext cwk_ token
• Each key is bound to a single organisation; all results are org-scoped and PII-redacted
• Read scopes cannot modify any data; action scopes can only test existing findings, never start a full pentest
• Keys can be revoked instantly in Settings > API Keys, and can be given an optional expiry
• Every request is rate-limited per key
• API-key creation and revocation are recorded in the Audit Log

Troubleshooting