forticnapp-mcp
Click on "Install Server".
Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state.
In the chat, type
@followed by the MCP server name and your instructions, e.g., "@forticnapp-mcpshow me the top 10 compliance findings"
That's it! The server will respond to your query, and you can continue using it as needed.
Here is a step-by-step guide with screenshots.
forticnapp-mcp
An MCP (Model Context Protocol) server that exposes FortiCNAPP (formerly Lacework) API 2.0
operations as typed, auth-aware tools. Tools aren't hand-written: they're generated at startup
from lw swagger.json (lw.yaml), so the tool surface tracks the spec you point it at.
Choose STDIO if you are building or running tools locally on your own machine. Choose Streamable HTTP if you need to share the server across multiple users, deploy it to the cloud, or require centralized authentication.
The same tool registry ships behind two interchangeable transports — pick whichever fits how you're running the server:
stdio ( | HTTP ( | |
Use case | One local client (Claude Desktop, Claude Code) launches the server as a subprocess | One shared server, multiple remote MCP clients over the network |
Transport | JSON-RPC over stdin/stdout | Streamable HTTP, |
Process model | Spawned per client, dies with it | Long-running, started once (bare metal or Docker) |
Auth | Inherits the FortiCNAPP credentials from its own env | Same FortiCNAPP credentials, plus a bearer token ( |
Tenancy | One FortiCNAPP account per subprocess | One FortiCNAPP account per running instance — every caller with the bearer token sees the same account, there's no per-client credential isolation |
Both modes share every other module (config.py, openapi_loader.py, auth.py,
http_client.py, tool_registry.py) — http_server.py only adds a Starlette app, a bearer-token
auth gate, and a /healthz endpoint in front of the same mcp.server.lowlevel.Server that
main.py runs over stdio.
Why this exists
FortiCNAPP's API 2.0 spec doesn't declare OpenAPI securitySchemes, has no operationId on any
of its 120+ operations, and mixes read-only "search" endpoints in with real mutations under the
same HTTP methods. This server works around all three: it hardcodes the real FortiCNAPP auth
handshake, derives deterministic tool names from method + path, and classifies "is this
mutating" from the path shape rather than the HTTP verb alone. See CLAUDE.md for the full list
of spec quirks this design accounts for.
Install
Requires Python 3.11+.
pip install -e ".[dev]"Configure
Quickest path — run the interactive setup command, which prompts for your credentials,
validates them against the real FortiCNAPP token endpoint, and writes .env, a portable
.mcp.json, and .gitignore:
forticnapp-mcp-setupOr configure by hand:
cp .env.example .env
# then fill in FORTICNAPP_API_BASE_URL, FORTICNAPP_KEY_ID, FORTICNAPP_API_SECRETCredentials come from the FortiCNAPP/Lacework console (Settings > API Keys), which issues a
keyId and a secret — both are required for the default auth mode. See
.env.example for every supported variable, including FORTICNAPP_ENABLED_TAGS (which OpenAPI
tags become tools) and ENABLE_MUTATION_TOOLS (off by default — only read-only tools are
exposed until you opt in).
Run: stdio mode (local, single client)
forticnapp-mcp
# equivalently:
python -m forticnapp_mcp.mainThe server speaks MCP over stdio — the client (e.g. Claude Desktop) launches it as a subprocess and talks JSON-RPC over its stdin/stdout. It validates configuration and loads the spec before it starts listening, and exits with a clear one-line error on stderr if either step fails — it will not start with a broken configuration. This is the default mode; see "Claude Desktop configuration" below for a full client config.
Run: HTTP mode (shared/remote, multiple clients)
For a shared/remote deployment (one server, multiple MCP clients over the network) instead of a
local stdio subprocess, use the forticnapp-mcp-http entrypoint. It speaks MCP over Streamable
HTTP and requires a bearer token (FORTICNAPP_MCP_HTTP_TOKEN) on every request to /mcp — this
is a separate secret from your FortiCNAPP credentials, protecting the HTTP endpoint itself. The
server still serves exactly one FortiCNAPP account per instance, same as stdio.
cp .env.example .env
# fill in FORTICNAPP_API_BASE_URL/KEY_ID/API_SECRET as usual, plus:
# FORTICNAPP_MCP_HTTP_TOKEN=<a long random secret>
forticnapp-mcp-http
# equivalently:
python -m forticnapp_mcp.http_serverOr run it in Docker instead of bare metal:
docker compose up --buildThe server listens on 0.0.0.0:8000 by default (FORTICNAPP_MCP_HTTP_HOST /
FORTICNAPP_MCP_HTTP_PORT). MCP clients should POST to http://<host>:8000/mcp/ with
Authorization: Bearer <FORTICNAPP_MCP_HTTP_TOKEN> (a bare /mcp without the trailing slash
307-redirects there). GET /healthz is unauthenticated and used by the container's HEALTHCHECK
and by any external load balancer/uptime check.
A client config pointing at HTTP mode looks like this (shape varies by client — this is the generic Streamable HTTP form):
{
"mcpServers": {
"forticnapp": {
"type": "http",
"url": "http://<host>:8000/mcp/",
"headers": { "Authorization": "Bearer <FORTICNAPP_MCP_HTTP_TOKEN>" }
}
}
}Develop
ruff check src/
pytestClaude Desktop configuration (stdio mode)
Add to claude_desktop_config.json:
{
"mcpServers": {
"forticnapp": {
"command": "python",
"args": ["-m", "forticnapp_mcp.main"],
"cwd": "/absolute/path/to/mcp_forticnapp",
"env": {
"FORTICNAPP_API_BASE_URL": "https://YourAccount.lacework.net",
"FORTICNAPP_KEY_ID": "YOUR_KEY_ID",
"FORTICNAPP_API_SECRET": "YOUR_SECRET",
"FORTICNAPP_OPENAPI_SPEC": "/absolute/path/to/mcp_forticnapp/lw.yaml"
}
}
}
}cwd must be the project root so the default FORTICNAPP_OPENAPI_SPEC=./lw.yaml resolves;
alternatively set an absolute path as shown above. Prefer a real .env file over inlining
secrets in this config where your setup allows it.
Architecture
config.py env vars -> validated Settings (pydantic-settings), fails fast
openapi_loader.py lw.yaml/json -> OperationSpec list (resolves $ref/allOf, builds
a pydantic input model per operation, infers the token endpoint's
field names)
auth.py ApiKeyAuthStrategy / BearerTokenStrategy / ApiKeyToTokenStrategy;
the last is FortiCNAPP's real keyId+secret -> bearer token handshake
http_client.py httpx.AsyncClient wrapper: builds requests from validated arguments,
retries network/5xx errors, retries once on 401 after refreshing auth,
follows FortiCNAPP's cursor-style pagination
tool_registry.py OperationSpec list -> mcp.types.Tool list (resolving any tool-name
collisions) and dispatches call_tool requests to http_client
main.py wires the above into mcp.server.lowlevel.Server + stdio_server
models.py OperationSpec/OperationParameter (internal) and the ToolCallResult/
RequestMeta/PaginationInfo pydantic models every tool returns
errors.py ForticnappError hierarchy (auth/validation/api/network/spec), each
carrying category/status_code/operation_id/retryable
logging_utils.py structured JSON logs to stderr with header/secret redaction
utils.py tool-name derivation, mutation/pagination classification, JSON
Schema -> Python type mappingEvery tool call returns the same structured JSON envelope:
{
"success": true,
"status_code": 200,
"operation_id": "forticnapp_alerts_list",
"request": {"method": "GET", "path": "/api/v2/Alerts", "query_keys": ["startTime"], "has_body": false},
"data": { "...": "..." },
"pagination": {"rows": 50, "total_rows": 400, "next_page_url": "https://...", "has_more": true},
"error": null
}To fetch the next page, call the same tool again with page_url set to
pagination.next_page_url from the previous response — every other argument is ignored when
page_url is set.
Customizing the token exchange
If your FortiCNAPP/Lacework deployment's token endpoint differs from the documented contract
(self-hosted, FedRAMP, a future API revision), there is exactly one place to change:
ApiKeyToTokenStrategy._acquire_token in auth.py. It builds the token request and parses the
response using field names from a TokenOperationHint that's inferred from the spec at startup
(openapi_loader.discover_token_operation) with fallback defaults matching FortiCNAPP's current
contract (keyId/expiryTime in, token/expiresAt out, secret carried in X-LW-UAKS).
Token caching, proactive refresh, and 401-triggered re-acquisition are all wire-format-agnostic
and live in the surrounding ApiKeyToTokenStrategy methods — you shouldn't need to touch them.
Security notes
Tokens and secrets are kept in memory only; nothing is persisted to disk.
logging_utils.redact_headers()/redact_secret()are used everywhere a header dict or secret reaches a log call —Authorization,X-LW-UAKS, and cookie headers are never logged in full.Mutating operations (anything that isn't a GET or a
POST .../search) are excluded from the tool list unlessENABLE_MUTATION_TOOLS=true.
This server cannot be installed
Resources
Unclaimed servers have limited discoverability.
Looking for Admin?
If you are the server author, to access and configure the admin panel.
Latest Blog Posts
- Your AI Chatbot Just Exposed Your CEO's Salary to an InternBy Om-Shree-0709 on .Agent IdentityMCP SecurityOAuth Delegation
- Why MCP Servers Need Execution Sandboxing (And Why Your Current Stack Isn't Enough)By Om-Shree-0709 on .Agentic AiPrompt InjectionWebAssembly
MCP directory API
We provide all the information about MCP servers via our MCP API.
curl -X GET 'https://glama.ai/api/mcp/v1/servers/svuillaume/forticnapp-mcp'
If you have feedback or need assistance with the MCP directory API, please join our Discord server