A Model Context Protocol server for Instagram API integration enabling AI applications to manage Instagram business accounts programmatically
This Instagram MCP server has mixed definition quality. Tool names follow verb_noun conventions well (get_, publish_, send_, validate_), and all tools have descriptions. However, parameter documentation is inconsistent, output schemas are undocumented, and error handling lacks recovery guidance. The schema for several tools is incomplete, parameter descriptions vary in quality from good to minimal. No tool annotations (readOnlyHint/destructiveHint) are present despite clear risk classifications. Pagination is implemented for some list tools but not consistently documented. Security parameter handling appears correct (no secrets exposed), but error responses do not guide LLM recovery.
Get account-level insights and analytics for Instagram business account
Get Facebook pages connected to the account and their Instagram business accounts
Get Instagram DM conversations. Requires instagram_manage_messages permission. Lists all conversations for the connected Instagram account.
Get detailed insights and analytics for a specific Instagram post
Get recent media posts from Instagram account with engagement metrics
Get Instagram business profile information including followers, bio, and account details
Output schemas not documented. LLMs cannot predict what fields tools return, forcing them to parse unstructured responses or make assumptions about downstream chaining. For example, publish_media, get_profile_info, and get_account_pages return structures that are inferred but not formally documented.
Tool annotations missing. Despite explicit Risk classifications (READ_ONLY, WRITE), no readOnlyHint, destructiveHint, or idempotentHint annotations are present in tool definitions. This prevents LLMs from understanding side effects without reading descriptions.
Error handling lacks recovery guidance. No evidence of actionable error messages or next-step hints. If an API call fails (e.g., invalid account_id, permission denied), the server does not guide the LLM to resolve the issue (e.g., 'Try validate_access_token() first' or 'Account not found. Available accounts: ...').
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | C | 63 | <=2025-11-25 | v2 |
| 2026-03-09 | F | 37 | - | v1 |
Upload and publish an image or video to Instagram with caption and optional location
Send a direct message to an Instagram user. Requires instagram_manage_messages permission.
Validate the Instagram API access token and check permissions
get_account_pages has minimal schema. The input object has no properties, description says 'Returns Facebook pages connected to the account' but parameter documentation is empty. LLMs cannot determine if this is a no-argument tool or if required params are missing.
Pagination documentation inconsistent. get_media_posts and get_conversations document 'after' cursor and 'limit' clearly, but the response structure (does it return a 'next_cursor', 'has_more', or implicit list termination?) is not documented. This forces LLMs to guess how to iterate.
validate_access_token has empty input schema. Input object has no properties, but description does not clarify if this is intentional (no parameters required) or an oversight. Clarity is needed.
publish_media uses anyOf for mutually exclusive parameters (image_url vs video_url). While structurally correct, the description should explicitly state 'Exactly one of image_url or video_url is required' and explain what happens if both or neither are provided.
Result limits not enforced or documented. get_media_posts and get_conversations default to 25 and max to 100, but there is no guidance on what happens if the agent requests 1000 items or calls these tools in a tight loop. No rate-limiting or runaway-protection hints in descriptions.