MCP server for creating, managing, and analyzing Meta (Facebook/Instagram) ad campaigns with Claude.
The Meta Ads Manager MCP server exhibits moderate definition quality with consistent but minimal descriptions and partial schema documentation. All 15 tools are explicitly defined with names, descriptions, and parameter schemas. However, descriptions are formulaic and lack depth (averaging ~50-80 chars, below the production baseline of 194 chars). Parameter descriptions are present but generic. Output schemas are not documented. Error handling guidance is absent. Tool naming is clear and follows verb_noun conventions, which is a strength. The server lacks enum constraints for multi-value parameters (e.g., 'status' accepts free-form strings rather than enums), inviting hallucinated values. No input validation rules, response field documentation, or chaining references are evident in the code.
Create an ad. Returns the ad ID.
Create an ad creative. Returns the creative ID.
Create an ad set with targeting. Returns the ad set ID.
Create an ad campaign. Returns the campaign ID.
Get configured ad account details.
Get ad sets for a campaign.
Get ads for a campaign.
Descriptions are formulaic and lack actionable depth. Most descriptions are 40-80 characters, well below the production baseline of 194 chars. For example, 'Get ad sets for a campaign' does not explain WHEN to call this vs list_ad_sets(), what fields are returned, or what the response structure looks like.
No output/return schema documentation visible in code. LLMs cannot infer what fields get_insights returns (impressions, clicks, spend, cpc, cpm, ctr, actions are mentioned in input description but not output). Similarly, create_campaign returns only 'campaign ID', what format? What fields are available for chaining into list_campaigns?
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 52 | 2026-07-28+ | v2 |
Get campaign details.
Get insights for an account, campaign, ad set, or ad.
List ad sets in the configured ad account.
List ads in the configured ad account.
List campaigns in the configured ad account.
Update daily budget for a campaign or ad set.
Update the status of a campaign, ad set, or ad.
Upload an ad image. Returns the image hash.
Parameter 'status' on update_status, create_campaign, create_ad_set, and create_ad accepts free-form strings with no enum constraint. Valid values appear to be 'PAUSED', 'ACTIVE', etc. (inferred from defaults), but LLMs will hallucinate invalid values like 'stopped', 'enabled', or 'running'.
Parameter descriptions lack validation rules and ranges. For example, 'daily_budget' in create_ad_set is typed as integer but lacks a description of units (cents? dollars?), minimum (1 cent?), or maximum. The code checks _check_daily_budget_limit() but this is not communicated to the LLM.
Error handling is silent. No tool descriptions mention what happens on failure (e.g., invalid campaign_id, rate limit, auth failure). The MetaAPIError class exists but error messages are not designed to guide LLM recovery. No retryable vs. fatal error classification.
Naming ambiguity: get_ad_sets vs get_ads (both return filtered results for a campaign) vs list_ad_sets/list_ads (return all in account). The distinction is not obvious from names alone. LLMs may confuse which to call when.
Mutually exclusive parameters not documented. create_ad_set accepts a 'targeting' object with age_min, age_max, genders, countries, interests, platforms, facebook_positions, instagram_positions, but it's unclear which are required vs. optional, and what happens if you omit countries (defaults to ['US']?).
No evidence of idempotency support or dry-run confirmation for destructive operations. create_campaign, create_ad, upload_image have side effects but no dry-run option or confirmation step to prevent accidental duplicate creation.
Fields parameter is a free-form comma-separated string with no schema or validation. If an LLM passes an invalid field name (e.g., 'campaign_status' instead of 'status'), the error response from Meta API is not guaranteed to be LLM-friendly.
No documentation of chaining IDs. If an LLM calls list_campaigns and wants to examine an ad set, what field in the campaign response contains campaign_id to pass to get_ad_sets? Responses are not explicitly documented, forcing the LLM to guess or infer.