Pikes MCP Server

Connect an MCP-compatible assistant to the Pikes creative suite: saved brand context, product and asset libraries, boards, editable projects, creative production, and campaign insights. Work starts with the brand’s own materials and stays organized in its Pikes workspace.

Pikes includes AI image generation and editing, video generation, and audio generation. Media generation spends the connected account’s credits; account permissions still apply to saved content. Meta analysis requires a separate Meta connection in Pikes.

https://pikes.ai/mcp

Protocol 2025-06-18 (also accepts 2024-11-05), Streamable HTTP with SSE for streaming responses. The server identifies itself as pikes-ai, titled Pikes, with PNG icons and a website URL so a client can show a proper connector card rather than a bare string, and advertises three capabilities — tools, prompts and resources, each with listChanged. The authenticated tools/list response supplies the current tool catalogue and input schemas; the API reference covers HTTP usage and errors.

Connect#

Claude (desktop or claude.ai)#

1. Install Pikes for Claude.
2. Review the prefilled name (Pikes) and remote MCP URL (https://pikes.ai/mcp), then continue.
3. Authorize when prompted — the OAuth screen signs you into Pikes and hands the connector a scoped token. No API key to paste, nothing to rotate by hand.

Claude Code / Cursor#

claude mcp add --transport http pikes https://pikes.ai/mcp

Or in .mcp.json:

{
  "mcpServers": {
    "pikes": {
      "type": "http",
      "url": "https://pikes.ai/mcp"
    }
  }
}

ChatGPT#

Add https://pikes.ai/mcp as a custom connector and authorize the same way. Select Pikes in the chat before running a creative workflow. For editable layouts, request an editable Pikes project, a preview and its project link. The preview card opens the saved project; board links preserve their existing visibility and require access to private boards.

Authentication#

OAuth 2.1 (preferred). Point the client at https://pikes.ai/mcp and it discovers everything it needs:

GET /.well-known/oauth-authorization-server
GET /.well-known/oauth-protected-resource

PKCE (S256) is required. CIMD clients may publish a set of supported token authentication methods; Pikes accepts the intersection with its advertised none and client_secret_post methods. A legacy singular preference does not override the plural capability set. Unsupported assertions are rejected, not ignored. Dynamic clients still use their registered authentication method.

Clients that register themselves do so through POST/mcp/oauth/register (RFC 7591); the rest of the flow runs through /mcp/oauth/authorize, /mcp/oauth/token and /mcp/oauth/revoke. Scopes: generate_image, edit_image, expand_image, read_images.

API key. For clients that can't complete OAuth, create a key under Settings → Claude / MCP in the app and send it as either header:

Authorization: Bearer psk_…
x-api-key: psk_…

A key carries the full rights of the user who made it, so treat it like a password: keys are stored only as a SHA-256 hash and can be revoked at any time from the same screen.

What the agent gets#

For a pasted asset/frame link, call read_item_link with { "url": "https://pikes.ai/app/projects/<project-id>?page=<page-id>&object=<object-id>" }. It returns the exact object and its current project/page location, including after a page move. Existing project access is required. Studio asset links (https://pikes.ai/s/<asset-id>) return the asset and project placements accessible to the connected user. A removed item returns an error.

The authenticated tools/list response is the source of truth for current parameters and defaults. The API reference provides HTTP examples. Meta tools need the user's Meta account connected in Pikes first.

Each tool ships a display title and the standard MCP annotations, so a client knows what it is dealing with before it calls. Read tools carry readOnlyHint; generation tools declare themselves as writes that are not idempotent — the same prompt twice is two images and two charges — and update_context and batch_rename_assets are marked destructiveHint, because both replace what was there. Tools that call generation providers, serve newly generated media through public URLs, or can change a public board or an already-public project advertise openWorldHint: true. The hint covers all supported modes, even when the particular request creates a private board. It does not grant sharing permission or change an object's visibility. Private project creation remains closed-world; renaming remains reversible.

Projects and canvas editing#

Native sections use add_project_object with objectType:"note", a title in content, dataJson:'{"canvasSection":true}', and bounds enclosing the member objects with padding. Membership is spatial and objects stay editable. To move a section with its contents, include the section, enclosed objects, and nested sections in layout_project_canvas with the same position delta.

Claude and other MCP clients use the same project operations as the Pikes agent and web canvas. Existing OAuth/API-key authentication applies; project access is limited to the caller's own projects and projects shared with their teams.

1. Call list_projects, then list_project_pages for the chosen project.
2. Read get_project_canvas with explicit projectId and pageId. It returns positions, dimensions, note text, editable frame layers, and asset URLs. Follow nextOffset until null when inspecting a whole page. This is structured canvas content, not a rendered screenshot.
3. Use layout_project_canvas for tidying; add_project_object for notes or frames; add_project_text to write on an image/frame; and update_project_text to edit one note or nested text layer. Text overlay creation preserves the underlying image and returns the new layerId. Note text edits also update the rich-text document used by the renderer, retaining paragraph/heading styles. Complex list documents become paragraphs on full-text replacement; use dataJson for precise document editing.
4. Canvas writes broadcast the saved objects to the private page channel so open browsers update without reload. A notification failure still reports the successful save with canvasRefreshRequired; refresh instead of repeating a write that already succeeded. Read the changed page back. Concurrent edits detected during text or frame updates return an error: reread before retrying. Layout batches can partially succeed if a database write fails; inspect before retrying.

All canvas writes require an explicit active page. Removing a canvas image preserves its library asset and records the existing sync exclusions. Removing a board also removes its linked canvas placements. Page trash is reversible. Whole-project deletion and sharing changes are not exposed by these tools.

Optional arguments can be omitted in MCP. dataJson, when supplied, is a JSON object encoded as a string and replaces the complete object data; preserve existing fields and layers. Frame layers paint bottom to top in the frame's native pixel coordinates. External MCP clients discover their project and page explicitly; they do not inherit the open browser tab.

After a server release, clients that cache discovery may need to refresh the connector or reconnect to receive the new tool list. No new database migration or credentials are required for these tools.

Prompts#

Four ready-made prompts, each taking arguments the client fills in:

Resources#

Once the connection is authorized, resources/list returns the account's own material plus the shared interactive result view:

The ui:// entries are MCP Apps (text/html;profile=mcp-app). Media creation returns the view directly. Read and intermediate write tools return complete structuredContent plus their existing text fallback, so agents can analyze and chain results without repeatedly opening an iframe. Call present_results with a source tool name and its result object to show the final selection. It supports every advertised source tool and never executes that tool again.

The app uses the standard MCP Apps bridge and a self-contained HTML resource (no external JavaScript dependency). It follows the host theme, reports its content height, loads media on demand, and uses explicit host capabilities for links, downloads, follow-up messages, and read-only job polling. A missing capability leaves a usable fallback. No button generates or mutates saved data automatically. A failed host action is shown inline, and cancellation stops job polling. See MCP app QA for the repeatable local check loop.

If a host caches the tool catalogue, refresh/reconnect Pikes once after upgrading to discover present_results. Existing resource URIs remain supported.

Discovery requests require sign-in#

initialize, tools/list and prompts/list all return 401 without a credential, carrying the challenge a client needs to build its sign-in link:

WWW-Authenticate: Bearer error="invalid_token",
  error_description="Authorization required",
  resource_metadata="https://pikes.ai/.well-known/oauth-protected-resource/mcp"

The JSON-RPC body carries the same thing in a form code can read — a stable data.code of auth_required plus absolute resource_metadata, authorization_endpoint and registration_endpoint URLs.

Protected Resource Metadata (RFC 9728) is served at every shape a client tries: /.well-known/oauth-protected-resource, the path-inserted /.well-known/oauth-protected-resource/mcp, and /mcp/.well-known/oauth-protected-resource. Authorization server metadata likewise answers at /.well-known/oauth-authorization-server, …/mcp, and /mcp/.well-known/oauth-authorization-server.

The JSON catalogue at GET/mcp (also /mcp/info) includes the same top-level resource, authorization_servers, and other protected resource metadata. Codex sign-in reads the configured MCP URL directly before following discovery; a catalogue without these fields fails with “Protected resource metadata missing required resource field” even when the .well-known endpoints are correct.

Calling a tool over plain HTTP#

An orchestration layer that doesn't want an MCP client can call the same tools directly — same auth, same arguments, JSON in and out:

curl -X POST https://pikes.ai/mcp/tools/generate_image \
  -H "x-api-key: psk_…" -H "Content-Type: application/json" \
  -d '{"prompt":"a matte black bottle on wet stone","aspectRatio":"4:5","resolution":"2K"}'

When something fails#

An authentication 401 from the /mcp JSON-RPC endpoint carries WWW-Authenticate and a data.code of auth_required; follow resource_metadata to the authorization server rather than hard-coding an endpoint.

Generations spend the connected account's credits, and a failed render is refunded. get_account_info reports the balance without leaving the agent.

More: API reference · Connecting an AI agent · MCP specification

Four creative-suite workflows#

These examples use the connected account’s own brand data and assets. Generation requires sufficient credits. Meta analysis additionally requires an authorized Meta connection; the other workflows do not.

Build a campaign from saved brand assets#

“Read our saved brand guidelines and palette, find our product packshots, and create a private Summer Launch board with the relevant assets. Tell me what is missing before creating new media.”

Expected result: a private board containing accessible existing assets, with saved brand context informing the selection. No new generation is needed.

Compose an editable campaign layout#

“Create a private Summer Launch project with a concept page. Arrange our saved product assets and a headline in an editable square layout. Preview the result and check for text overflow.”

compose_project_frame creates native image, text, and graphic layers from a layer plan and prepared assets. preview_project_frame renders the actual saved frame and returns a preview URL, overflow measurements, and a visual critique. The saved layers remain editable. A structural success alone does not establish visual quality.

Produce and organize campaign media#

“Using the selected packshot and saved brand colors, make two lifestyle-image variations. After I choose one, animate it into a short product video. Save the outputs and add them to our Summer Launch board. Confirm the creative direction and credit-using steps with me first.”

Expected result: saved image and video assets linked to the campaign board. Generation spends Pikes credits. A pending job or failed render is not a completed asset; completed results contain the saved asset IDs and media URLs.

Use campaign performance to inform the next brief#

“Compare our connected Meta ad creatives over the last 30 days by spend, purchases, and ROAS. Identify supported patterns and propose three concepts using our saved brand context. Do not publish ads, change budgets, or generate assets.”

Expected result: observations tied to the requested dates, account currency, and retrieved metrics, plus a proposed brief. Pagination limits and unavailable metrics should remain explicit. Competitor-ad longevity is not proof of returns. The connector does not publish ads or change campaign budgets.

Troubleshooting and review setup#

• Sign-in or 401 response: reconnect and complete OAuth. A session ID is a routing identifier, not an access credential. Protected calls need a current bearer token or API key.
• Permission denied: confirm the account has access to the project, page, board, or asset. IDs do not confer access to another account’s data.
• No saved products or assets: import or upload appropriate source material; an empty library response does not mean new media has been created.
• Insufficient credits: check the account balance before requesting new generation. Reading assets and preparing a brief need no media generation.
• Meta not connected: connect an authorized Meta account in Pikes for ad analysis. Brand-library and design workflows remain separate.
• Rendering failure or partial layout update: inspect the reported job or saved page before retrying. A repeated generation request may incur another charge; a layout batch may already have applied some changes.

A reviewer workspace should contain a fictional brand brief, an exact color palette, product images, an asset folder, a private board, and an editable project with text and image layers. Full media testing also requires credits. Meta review requires a separately authorized test account with real accessible metrics; synthetic metrics must not be presented as live Meta results.

Support: leo@pikes.ai. Data handling and retention: Privacy policy.