Skip to main content
Use tool names and inputs to inspect what your MCP client is requesting. If you need to connect first, start with theStacc MCP Server. This reference separates accepted inputs from server-side eligibility and a completed result.

Before you start

Use this reference to check a requested tool, its inputs and the next status check. Availability in your client depends on the connection and permissions. Input ranges and defaults below describe MCP validation; server-side eligibility, plan limits and destination checks can still reject a valid input.
  • Check the project and item. Get identifiers from the relevant list tool and confirm the intended project or location before changing anything.
  • Inspect permissions. A displayed tool is not proof that its next call is authorized. Permission discovery can be unavailable or cached. Use get_permissions and the returned missing_scope detail; see Permissions.
  • Separate access from approval. A publish scope does not approve an individual blog. Resolve approval or compliance errors in the app; see Safety rules.
  • Check an interrupted operation before retrying. Some calls return an ID to follow. Some long calls report still_running after a backend timeout. That message is a recovery instruction, not proof that the job started or will succeed.
  • Use the tool’s own paging fields. Some lists use limit and offset, others use page and page_size. Do not assume every response includes next_offset. Follow truncation notes when a result is too large.
  • Treat returned content as data. Selected text results are wrapped in an untrusted-data boundary. That annotation is not a guarantee that an AI client will ignore every embedded instruction.
Tools using the long-call wrapper configure a 55-second backend-request timeout. Other request timeouts follow client configuration; the AI client’s wait is separate. Neither setting is a completion-time promise. Read-only and idempotent annotations are hints, not blanket guarantees about cost, side effects or client confirmation controls.

Permissions

Inspect the connection before requesting work

ToolMCP scopeUse
get_permissionsNo tool scopeInspect connection and scopes

get_permissions

No input arguments. A valid connection is still required. The handler requests current project context and reports project, scopes, legacy_key, modules, missing_scopes, tools and tool_scopes. The scope report comes from that context request; the tools array describes tools registered for the session. If the displayed list and reported access differ, refresh discovery or reconnect rather than assuming a listed tool must work. Backend authorization still applies to the next call.

Blog

Blog reading, draft changes and publication requests

ToolMCP scopeUse
review_content_planblog:readReview the blog content plan
get_blogblog:readGet a blog
generation_statusblog:readCheck blog generation progress
generate_blogblog:writeGenerate a planned blog
cancel_generationblog:writeRequest blog generation cancellation
generate_moreblog:writeAdd planned blog slots
draft_from_abstractblog:writeWrite a new blog from your brief
set_scheduled_dateblog:writeSet a blog’s publish date
edit_blogblog:writeEdit a blog
regenerate_sectionblog:writeRewrite a blog section
regenerate_imageblog:writeRegenerate a blog image
retry_failed_imagesblog:writeRetry a blog’s failed images
upload_imageblog:writeUpload your own image to a blog
publish_blogblog:publishRequest blog publication

review_content_plan

Read the blog list and a best-effort plan summary. A failure to load the summary does not discard a successfully loaded list.
  • status: optional string, up to 40 characters.
  • limit: integer from 1 to 200, default 50.
  • offset: non-negative integer, default 0.
The wrapper reports blogs, total, offset and next_offset. Continue with the returned offset when there is another page.

generate_blog

Request generation for an existing planned blog.
  • blog_id: required UUID.
  • additional_direction: optional, up to 2,000 characters.
Read the returned state and follow generation_status. If generation is refused because the blog is awaiting a date, use set_scheduled_date for that state. Do not interpret a lost response as permission to start another generation.

generation_status

blog_id is a required UUID. Use this read to inspect the reported generation state and any available progress or error details. A failed state needs investigation; polling does not guarantee that the blog will eventually reach review.

cancel_generation

blog_id is a required UUID. This sends a cancellation request for the blog’s generation. Read the response and inspect the resulting blog with get_blog; do not assume an acknowledgement proves every background step stopped immediately.

generate_more

count is a required integer from 1 to 10. The tool requests additional planned blogs for the current project. Inspect the returned slots and plan limits; adding a slot is separate from completing its topic, assigning a date and generating its article. See Generate more.

draft_from_abstract

Create a blog request from your own brief.
  • abstract: required, from 10 to 4,000 characters.
  • blog_type: optional string, such as how_to.
When the response identifies a new blog, follow it with generation_status. After a timeout or unclear response, check review_content_plan before calling again: a repeated creation request can create another item.

set_scheduled_date

Assign a date to a blog in awaiting_user_date.
  • blog_id: required UUID.
  • scheduled_date: required YYYY-MM-DD string.
The date endpoint checks a real calendar date. It accepts today or a future date; yesterday in UTC is coerced to today as a time-zone grace rule. Other past dates are refused. If topic planning is unfinished, the endpoint attempts planning before committing the date and status change. Read any planning or conflict error before retrying. This tool is not a general rescheduler for every blog state.

get_blog

  • blog_id: required UUID.
  • include_config: boolean, default false.
Read the latest content before editing. By default, the wrapper keeps image-regeneration fields but removes other generator working data from ai_generation_config. Set include_config to true only when you need the full returned configuration.

edit_blog

  • blog_id: required UUID.
  • title, content, excerpt, meta_title, meta_description: optional strings.
  • allow_image_removal: optional boolean, for an intentional removal.
Only supplied edit fields are sent. Keep the images and content you intend to retain, then verify the saved result. The blog edit endpoint can refuse silent image removal. This tool’s input schema does not expose an expected-version argument, so do not rely on a universal concurrent-edit safeguard. Review current approval and sync requirements after changing a draft or published article; see Blog Status Lifecycle.

regenerate_section

Rewrite a supplied HTML section.
  • blog_id: required UUID.
  • section_html: required, up to 100,000 characters.
  • tone and length: integers from 0 to 100, default 50.
  • additional_prompt: optional, up to 1,000 characters.
  • persist: boolean, default true. Set false for a rewrite without requesting a save.
Check success and persisted. A successful rewrite is not necessarily a saved edit. With persistence requested, the endpoint reports false when the original snippet no longer matches the blog. Re-read it before placing the returned HTML; do not replace unrelated content to force a match.

regenerate_image

Request a replacement for a particular blog image.
  • blog_id: required UUID.
  • image_src: required, 1 to 2,000 characters, matching the current image source.
  • image_alt: required, up to 500 characters.
  • style: integer from 0 to 100, default 50.
  • custom_prompt: optional, up to 1,000 characters.
  • mode: optional create or edit. A non-empty prompt requires a mode; edit requires a non-empty prompt.
  • image_type: optional photo, diagram, infographic, illustration or brand_text.
  • image_occurrence: optional integer from 0 to 100, paired with image_occurrence_of, an integer from 1 to 100.
Follow the image-regeneration fields returned by get_blog, such as image_regen_status, image_regen_new_url and image_regen_error. Verify the chosen image in the content before publishing. Use the reported image state to decide the next step.

retry_failed_images

blog_id is a required UUID. Request another attempt for failed blog images, then inspect the returned details and blog state. This handler uses the 55-second long-call request setting. If it reports still_running, check get_blog before repeating the request. A timeout does not establish which images, if any, were retried. Inspect the current blog before another attempt.

upload_image

Upload a supplied image and request its placement.
  • blog_id: required UUID.
  • image_base64 with filename: inline image data. A filename is required for this input path.
  • file_path: optional local source, available only when local file reads are enabled for the MCP process. A hosted server cannot use a path on your laptop.
  • alt_text: optional, up to 500 characters.
  • placement: replace, append, featured or none, default replace.
  • image_index: non-negative integer, default 0, used for replacement.
The hosted server limits the whole request body to 8 MiB. Base64 data and the JSON envelope both consume that space. The blog destination also validates decoded image data and applies its own asset limit; respect both the transport body limit and the destination’s asset limit. Check the returned asset and placement rather than treating storage as proof the image appears correctly in the public page.

publish_blog

  • blog_id: required UUID.
  • integration_id: optional string; otherwise the service selects the connection.
The blog publishing gate checks human approval state and approval validity for non-human principals. Resolve not_approved, approved_by_assistant or approval_stale by reviewing the current draft in the app. Publication can also be refused by SEO or compliance checks; use the returned code and details instead of assuming an HTTP success means publication succeeded. This handler uses the long-call wrapper with get_blog as its recovery read. Inspect the returned state and destination. A response is not a guarantee that a public URL is available or that media and page content are correct. See the publishing guide.

SEO and analytics

Project analytics and SEO reporting reads

ToolMCP scopeUse
get_analytics_statusseo:readCheck Search Console and GA4 connection
get_search_console_performanceseo:readSearch Console performance
get_seo_insightsseo:readSEO insights and next actions
get_ga4_overviewseo:readGA4 traffic overview
get_blog_performanceseo:readPer-blog performance
get_monthly_seo_reportseo:readMonthly SEO report
Reporting completeness depends on connected properties and available data. Check get_analytics_status when a report is unavailable; do not infer a connection problem from every not_found error.
  • get_search_console_performance and get_ga4_overview: days, integer from 1 to 180, default 28.
  • get_blog_performance: range, either 28d or 90d, default 28d.
  • get_monthly_seo_report: optional month matching YYYY-MM, with a month from 01 to 12.
  • get_analytics_status and get_seo_insights: no input arguments.
Read the reported dates and synchronization state before interpreting the numbers. Available reporting periods and synchronization state matter more than an assumed refresh interval.

Site audit

Audit requests and result reads

ToolMCP scopeUse
run_site_auditseo:writeRequest an audit run
get_audit_statusseo:readSite audit progress
get_audit_dashboardseo:readSite audit summary
get_audit_issuesseo:readSite audit findings
get_audit_findingseo:readOne site audit finding
get_audit_pagesseo:readCrawled pages from the site audit

run_site_audit

No input arguments. Request an audit for the selected project. Check the latest audit state first and inspect any allowance or eligibility error. If the response is lost, use get_audit_status before starting another run. Follow the reported run to a completed or failed outcome rather than repeatedly submitting the start request.

Reading results

  • get_audit_status: optional run_id UUID. Without it, the wrapper requests the most recent run; it can report that no run exists.
  • get_audit_dashboard: no input arguments.
  • get_audit_issues: optional stage from availability, indexability, discoverability, page_understanding, search_relevance, experience, ai_retrieval; severity from critical, warning, info; status from open, regressed, verifying, resolved; and search up to 200 characters. page is an integer from 1, default 1; page_size is 1 to 200, default 50.
  • get_audit_finding: required finding_id UUID.
  • get_audit_pages: optional disposition string, up to 40 characters; limit is 1 to 500, default 50. This schema has no offset field.
Use the relevant finding or page read for details. Do not assume every paged tool uses the same response shape.

Keywords

Keyword-pool reads and changes

ToolMCP scopeUse
list_keywordsseo:readList keywords
add_keywordskeywords:writeAdd keywords
generate_keywordskeywords:writeResearch new keywords
  • list_keywords: limit from 1 to 500, default 100; non-negative offset, default 0. The wrapper slices the returned keyword pool and reports total, offset and next_offset.
  • add_keywords: required keywords array with 1 to 50 strings, each 1 to 80 characters.
  • generate_keywords: no input arguments.
Inspect the response for duplicates, rejected entries and plan limits. Adding or researching keywords does not establish that an article was scheduled, generated or ranked.

Social

Social post reads, edits and publication requests

ToolMCP scopeUse
social_list_postssocial:readList social posts
social_get_postsocial:readGet a social post
social_generation_statussocial:readCheck social post generation progress
social_generate_postsocial:writeCreate a social post draft
social_edit_postsocial:writeEdit a social post
social_regenerate_captionsocial:writeRewrite a social caption
social_regenerate_imagesocial:writeRedraw a social post’s image
social_edit_imagesocial:writeEdit a social image with an instruction
social_list_image_versionssocial:readList a social post’s image versions
social_use_image_versionsocial:writeUse a social image version
social_publishsocial:publishRequest social publication or scheduling
Read the post before sending an edit. Resolve approval, status or content-removal errors in the app or returned workflow; a write scope is not permission to ignore those checks.

social_list_posts

  • status: optional, up to 40 characters.
  • content_plan_id: optional UUID.
  • limit: integer from 1 to 500, default 25.
  • offset: non-negative integer, default 0.
The wrapper returns the service’s list response. Check its total and paging information rather than assuming it has the blog list’s next_offset field.

social_generate_post

Choose one request path:
  • New post: topic, 3 to 500 characters. Optional platforms is a non-empty array of linkedin, instagram, x or facebook; optional scheduled_date is YYYY-MM-DD.
  • Existing planned post: post_id UUID. Do not also send topic, platforms or scheduled_date.
  • Either path: optional additional_direction, up to 1,000 characters.
The existing-post path sends only_if_unwritten; it is not a request to overwrite a finished draft. On the new-post path, the wrapper first adds a topic and then requests generation. If that second request fails, its error includes the created post’s ID when available. Inspect that item instead of creating it again. Follow social_generation_status, then read the post; there is no guaranteed generation duration.

social_get_post and social_generation_status

Both require post_id as a UUID. Use social_get_post for the current content and social_generation_status for the reported generation or image state. Read any failure details before another write request.

social_edit_post

  • post_id: required UUID.
  • caption: optional, up to 10,000 characters.
  • platform_captions and first_comment: optional maps of string keys to string values.
  • hashtags: optional array of up to 60 strings, each up to 100 characters.
  • scheduled_date: optional YYYY-MM-DD.
  • scheduled_time: optional local-time string matching hours and minutes, with optional seconds.
  • allow_content_removal: optional boolean for an intended removal.
The wrapper forwards only these edit fields. Preserve all platform entries you still need rather than assuming a partial map will be merged. This schema has no status or approval field. Verify the saved content and schedule after the request.

social_regenerate_caption

  • post_id: required UUID.
  • platform: required linkedin, instagram, x or facebook.
  • additional_direction: optional, up to 500 characters.
Request a rewrite for that platform, then read the saved post and any error. Check the updated post before another rewrite.

social_regenerate_image

  • post_id: required UUID.
  • in_image_copy: optional object. headline, mechanism_line and outcome_line are each optional strings up to 300 characters; cta_text is up to 100.
  • slide_content: optional array of up to 12 slide objects.
This handler uses the long-call wrapper and names social_generation_status for recovery. Check image state, then inspect the post. A delayed response does not prove the redraw finished or that another request would be harmless.

social_edit_image

  • post_id: required UUID.
  • instruction: required, 1 to 4,096 characters.
  • intent: optional scene or image.
  • slide_index: optional integer from 0 to 11; apply_all: optional boolean.
  • request_id: optional UUID.
Keep the same explicit request_id when retrying the same intended edit. Without one, the wrapper tries to derive an ID from the post, current image revision and arguments. If that revision read fails, it falls back to a random ID. Do not treat retries as universally free or exactly once. Check social_list_image_versions for the edit job and candidate version before requesting another edit. This handler sets a 55-second request deadline but does not use the still_running conversion wrapper. An upstream_timeout needs a state check too.

social_use_image_version

  • post_id: required UUID.
  • revision_id: required string, 1 to 100 characters.
  • expected_updated_at: optional string, up to 80 characters.
If the expected version is omitted, the wrapper first reads the current image-version data and supplies its updated_at when available. Resolve a conflict by re-reading the item rather than forcing an older version over a newer change. Verify the selected image and review state afterward.

social_publish

  • post_id: required UUID.
  • publish_now: required boolean.
  • scheduled_for: optional string, up to 40 characters, for the requested scheduling time.
  • timezone: optional string, up to 64 characters.
  • platforms: optional non-empty array of linkedin, instagram, x or facebook.
The request is subject to the service’s approval, status, destination and publication checks. This handler uses the long-call wrapper with social_get_post for recovery. Inspect both the returned state and intended platforms; a scheduling acknowledgement is not proof the post appeared publicly.

Google Business

Location-specific Google Business Profile tools

ToolMCP scopeUse
gbp_list_locationsgbp:readList Google Business Profile locations
gbp_get_locationgbp:readGet a GBP location
gbp_connection_statusgbp:readCheck a GBP location’s connection
gbp_list_postsgbp:readList GBP posts
gbp_get_postgbp:readGet a GBP post
gbp_content_plangbp:readGBP content plan
gbp_generate_postgbp:writeCreate a GBP post draft
gbp_regenerate_postgbp:writeRewrite a GBP post
gbp_edit_postgbp:writeEdit a GBP post
gbp_generate_imagegbp:writeGenerate a GBP post image
gbp_upload_imagegbp:writeUpload your own GBP post image
gbp_remove_imagegbp:writeRemove a GBP post’s image
gbp_publishgbp:publishRequest publication to Google
gbp_retry_publishgbp:publishRetry a failed publication
gbp_publish_statusgbp:readRequest a publication-status refresh
Start with gbp_list_locations, which supplies the connection’s project ID when listing locations. Other GBP tools require location_id as a UUID; tools acting on a post also require a post_id UUID. Read the intended location and post before changing anything.

Reading

  • gbp_get_location, gbp_connection_status and gbp_content_plan: location_id UUID.
  • gbp_list_posts: location_id; optional integer year from 2000 to 2100, month from 1 to 12, and status up to 40 characters. limit is 1 to 100, default 50; offset is non-negative, default 0.
  • gbp_get_post and gbp_publish_status: both location and post UUIDs.
gbp_publish_status requests a status refresh through the service’s sync-status endpoint. Check its reported result. A read-style tool annotation is not a universal promise that no stored state is updated.

gbp_generate_post

  • location_id: required UUID.
  • post_type: required update, offer or event.
  • scheduled_date: required YYYY-MM-DD.
  • topic: optional, up to 500 characters.
This creates a post request and uses the long-call wrapper with gbp_list_posts for recovery. Inspect the list after an unclear or timed-out response before creating another post.

gbp_regenerate_post

Supply the location and post UUIDs, with optional user_direction up to 500 characters. The long-call wrapper names gbp_get_post for recovery. Check the current content, status and errors before repeating the rewrite.

gbp_edit_post

Supply the location and post UUIDs, then only the fields you intend to change:
  • title: up to 300 characters; content: up to 1,500.
  • cta_type: string up to 20 characters; choose the service’s supported action, such as learn_more or book.
  • cta_url: string up to 2,000 characters; use an appropriate secure destination.
  • offer_code: up to 100 characters; offer_terms: up to 5,000.
  • scheduled_date: YYYY-MM-DD; scheduled_time: string up to 5 characters, intended as local HH:MM. An empty value requests the location’s default time.
  • allow_content_removal: optional boolean for an intended removal.
The MCP schema’s text limit is not a complete Google policy check. The wrapper forwards only these fields and does not expose a status or approval argument. Inspect validation and review requirements in the response.

Images

All three image tools require the location and post UUIDs.
  • gbp_generate_image: optional additional_direction, up to 500 characters. The long-call wrapper names gbp_get_post for recovery.
  • gbp_upload_image: inline image_base64 with filename, or a supported local file_path. The wrapper rejects decoded data above 5 MiB. It maps JPEG/PNG filename extensions to their MIME types; the receiving service still validates the image.
  • gbp_remove_image: allow_content_removal defaults to false. Set it to true only when removal is intended, and inspect the result.
Hosted uploads also share the whole-request 8 MiB body limit. Check the resulting image and content; a size check is not a guarantee of format acceptance or successful publication.

gbp_publish and gbp_retry_publish

Both require the location and post UUIDs. gbp_publish requests publication; gbp_retry_publish requests another attempt after a failed publication. Both use the long-call wrapper and name gbp_publish_status for recovery. Resolve returned approval, status or connection errors before retrying. Verify the destination and final publication state instead of treating publishing as a completed result.

Tools that run in the background

Use the read appropriate to the operation. A status tool can report failure as well as progress; do not poll indefinitely or start another paid request merely because a response was lost.

Reads to check before repeating an operation

OperationRecovery readWhat to inspect
generate_blog, draft_from_abstractgeneration_statusReported blog state; check the plan for an unclear creation result.
regenerate_image, retry_failed_images, publish_blogget_blogImage state, saved content or publication result.
social_generate_post, social_regenerate_imagesocial_generation_statusGeneration or image state, then read the post.
social_edit_imagesocial_list_image_versionsJob and candidate version before another edit.
social_publishsocial_get_postRequested platforms and reported publication state.
gbp_generate_postgbp_list_postsWhether a post was created before another create request.
gbp_regenerate_post, gbp_generate_imagegbp_get_postCurrent content, image and reported errors.
gbp_publish, gbp_retry_publishgbp_publish_statusPublication state after the request.
run_site_auditget_audit_statusThe latest or identified run, including any failure.
When a response supplies check_with, follow that read and inspect the actual state. The MCP’s timeout-to-still_running conversion does not establish that the backend admitted or completed a job.

A typical session

  1. Use get_permissions to inspect the project and reported access.
  2. List the relevant blogs, social posts or locations, then read the intended item.
  3. Request one bounded change and inspect the saved result.
  4. Review current content and approval requirements in theStacc before publication.
  5. After a publication request, check its state and the actual public destination.
These steps are a review workflow, not a guarantee that every client asks for confirmation before each edit.

Guided workflows

The MCP also defines prompts. Client support determines how they are presented; they are not scheduled automations or guarantees that every requested step will complete. When permissions are known, prompt listing uses the scope checks below. If discovery is unavailable, a prompt can still be listed; authorization applies when its tools run.
  • weekly_content_review: offered when Blog, Social or GBP read access is present.
  • seo_next_actions: requires seo:read.
  • create_social_post: requires both social:read and social:write; topic is required and date is optional. Underlying tools still apply their own input limits.
  • refresh_declining_post: requires seo:read, blog:read and blog:write.
Inspect the prompt’s request before running it. See Guided workflows for the connection guide’s workflow context.