Integration · Contentful

Headless CMS with structured content models, both flavors of MCP available.

Contentful's first-party MCP server ships in two forms: a local open-source option (@contentful/mcp-server, MIT) and a hosted beta at mcp.contentful.com/mcp. Pick the deployment shape that fits the team's posture.

The local MCP runs in the agent runtime with a Management API token; teams that want full control over auth keep this. The hosted beta runs OAuth-handshake at Contentful's edge; teams that want zero infrastructure use this. Both surface the same tool set: entries and assets CRUD, content type management, locale handling, AI Actions, and the full publishing workflow with environments and validations. The structured content model is where Contentful's strength shows when the MCP is doing the writing: content types with field validations, references between types, and per-locale content the agent has to respect on every payload.

Official MCPContentful local MCP (MIT) plus hosted beta at mcp.contentful.com/mcp.

What you can do via MCP

Example prompts the agent runs.

  • “Create a new entry in the Article content type with the title, slug, body, and SEO fields filled out, in the staging environment.”

    Calls create-entry with the content-type ID, the field payload validated against the content type's schema, and the staging environment scope. Returns the entry ID; publish is a separate step.

  • “Update the German locale of the homepage hero with this translated headline and subhead; leave the English version alone.”

    Targets the entry by slug, updates only the de-DE locale fields, returns the diff so the editor can confirm before publishing the localized variant.

  • “List the content types in this space and show which ones have fields with missing validations.”

    Walks the content-type definitions, returns a structured list with content types, fields, and the validation gaps. Useful for a content-model audit before a migration.

  • “Run the AI Action tagged 'expand-summary' on this entry's draft body and produce a long-form version.”

    Calls invoke-ai-action with the AI Action slug, passes the entry's draft body as input, returns the generated long-form output for the editor to review and accept into a new field.

  • “Publish all entries in the Article content type that have been in 'changed' state for more than 7 days; show me the list first.”

    Filters entries by state and modified date, returns the candidate list for confirmation, then calls publish-entry on the confirmed set. Two-step pattern keeps stale-content publishes auditable.

Contentful · MCP

Entry creation via the Contentful MCP. The agent shapes a payload that matches the content type's field schema and validations.

Contentful MCP
docs
# Contentful MCP, called via the agent runtime
mcp.contentful.create_entry({
  space_id: "abc123",
  environment_id: "staging",
  content_type_id: "article",
  fields: {
    title: { "en-US": "How structured CMS content survives the brief" },
    slug: { "en-US": "structured-cms-content-survives-the-brief" },
    body: { "en-US": "<rich-text>...</rich-text>" },
    seo_title: { "en-US": "Structured CMS content | RampStack" },
    seo_description: { "en-US": "Why content models keep marketing output coherent." },
    primary_author: {
      "en-US": { sys: { type: "Link", linkType: "Entry", id: "author_aml" } }
    }
  }
})

# Returns the entry sys.id and version. Publishing is a separate call.
One command sample showing how the agent talks to Contentful. The MCP exposes the platform's primitives as tools; the agent translates the prompt into the right call.

MCP integration

Contentful MCP server.

Server
https://mcp.contentful.com/mcp (hosted beta) or @contentful/mcp-server (local, MIT)
Auth
Hosted: OAuth (Contentful). Local: Management API token (CMA token, scoped per space).
Hosting
Hosted beta on Contentful edge, or local in agent runtime. Pick by posture.
Scoping note
Hosted beta scopes to the spaces and environments the OAuth user can access. Local MCP scopes to whatever the CMA token allows; rotate tokens like any other secret.
  • Two deployment shapes: local open-source (MIT) and hosted beta (OAuth)
  • Entries and assets CRUD with content-type-aware validation
  • Locale-scoped editing for multi-language content models
  • AI Actions invocation: pre-built content workflows the team configured in Contentful
  • Environment-aware publishing: staging, master, and custom environments
  • Strong fit for structured content models with cross-references and per-locale fields

Contentful MCP docs

Visual demonstration

What this looks like in practice.

Contentful· CMS schema/ acme-marketing3 collections

  • Article

    Long-form articles with multi-locale support and reference fields. The Article content type carries the brief-axis tag for queryable filtering.

    412 entries
    • titleSymbolrequiredlocalized
    • slugSymbolrequiredunique
    • bodyRich textrequiredlocalized
    • seo_titleSymbollocalized
    • seo_descriptionSymbollocalized
    • hero_imageAssetrequired
    • primary_authorReference
    • brief_axisSymbolvalidated

  • Page

    Composable landing pages built from referenced sections. Each section is a separate content type referenced via a multi-link field.

    58 entries
    • titleSymbolrequired
    • slugSymbolrequiredunique
    • sectionsReferencevalidated
    • seo_titleSymbollocalized
    • seo_descriptionSymbollocalized

  • Author

    Author records. Referenced by Article via primary_author. One record per author across all locales.

    21 entries
    • nameSymbolrequired
    • slugSymbolrequiredunique
    • headshotAsset
    • bioRich textlocalized
The content model the Contentful MCP exposes. Localized fields, cross-content-type references, and field-level validations the agent has to respect on every write. The schema is the contract; the MCP enforces it.

CLI alternative

Contentful CLI for migrations and CI workflows.

Contentful ships a CLI for content-model migrations, environment promotion, and bulk imports. The MCP is the primary surface for AI-driven content editing; the CLI is the right shape for migration scripts that run unattended in CI and for one-time bulk operations the agent runtime is not built for.

Pairs with these skills

The content stack skill suite.

This integration pairs with the forthcoming content-strategy-for-ai-search and content-production skills. The Contentful microsite assumes the content model and the editorial brief exist; the integration shows where the published surface lives. Skill pages and SKILL.md sources land in subsequent dispatches; cross-link hyperlinks are added when the skill pages ship.

For teams that prioritize structured content models, cross-channel delivery, and per-locale workflows, Contentful is the canonical pick. For design-led marketing sites with strong ownership of visual layout, Webflow is the parallel choice; both ship official MCPs.

Continue reading.