MCP server for automating and managing Requestly rules, groups, and integrations via the Requestly API.
The server defines 8 tools with mostly complete schemas and descriptions. Naming follows verb_noun conventions (create_rule, get_rules, etc.). Descriptions are present and substantial (100-300+ chars), exceeding the 10-1024 baseline. However, there are gaps in parameter descriptions, inconsistent schema presentation, and missing output schema documentation. Destructive operations (delete_rule, delete_group) correctly implement confirmation gates. Tool annotations are present (toolAnnotations=true). The schema quality varies: some tools show full Zod definitions (createGroup.ts visible), but others are inferred from the tool definitions table without visible source validation. Error handling is present but generic in places.
Create a new group in Requestly.
This endpoint allows you to create various types of rules in Requestly, such as Redirect, Cancel, Replace, Headers, User-Agent, Script (Insert Script), Query Param, Modify Request, Modify Response, and Delay. Each rule has a specific structure and parameters based on the ruleType. New rules default to Inactive. Script/Request/Response rules (which can execute code in the browser/request pipeline) are ALWAYS created Inactive and must be reviewed by a human and explicitly activated before they take effect.
Delete a specific group in Requestly using its id. IRREVERSIBLE and CASCADES — deleting a group also deletes every rule it contains. Requires confirm: true. Never set confirm: true on the user's behalf — first show the user the exact group id and get their explicit approval.
Delete a specific rule in Requestly using its ruleId. IRREVERSIBLE. Requires confirm: true. Never set confirm: true on the user's behalf — first show the user the exact ruleId and get their explicit approval.
Get all groups in Requestly.
Output schemas not documented. Tool descriptions state what the tool does but do not explain the structure of returned data. LLMs cannot plan downstream calls without knowing what fields to expect (e.g., does get_rules return rule_id, ruleId, or id?).
Parameter descriptions incomplete or generic for several tools. get_rules and get_groups lack detail on pagination behavior, result count, and field structure. For example, 'pageSize' states 'max 75' but does not explain default behavior or what happens if omitted.
Pairs parameter in create_rule and update_rule lacks detailed schema documentation. Description states 'List of rule pair objects. Structure depends on ruleType' but does not define the exact structure for each rule type (Redirect, Cancel, Replace, Headers, etc.). This forces LLMs to guess the schema.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | C | 67 | 2026-07-28+ | v2 |
| 2026-03-09 | F | 48 | - | v1 |
Retrieve all rules or a specific rule from Requestly using its API. Supports pagination and lookup by ruleId.
Update a specific group in Requestly.
Update an existing rule in Requestly. Requires ruleId and the updated rule payload. Activating a Script/Request/Response rule (status: 'Active') runs code in the browser/request pipeline and requires confirmActivation: true — never set it on the user's behalf; a human must review the rule's script/code first.
create_rule and update_rule descriptions mention that Script/Request/Response rules must be reviewed and activated by humans, but the confirmActivation parameter is only on update_rule, not create_rule. This inconsistency may confuse LLMs about which tool requires explicit confirmation.
API key injection via environment variable (REQUESTLY_API_KEY) is correct, but there is no validation that the key is present or valid before attempting API calls. apiClient.ts includes security patterns (RQ-3018 traversal defense, RQ-3025 body redaction) but error messages for missing credentials are generic.