How to Connect Amp to Social Media With MCP

Connect Amp to social media by adding Groniz as a remote MCP server with amp mcp add or the documented amp.mcpServers configuration. Point it to https://mcp.groniz.com/mcp. Decide where the setup must apply because Amp resolves MCP configuration in this order: CLI flags, workspace, user, then skills. A workspace server also requires approval.

Choose whether the action runs locally or inside an Orb. Local MCP settings do not automatically enter Orbs, and browser OAuth is unavailable inside them. Give Amp an approved source asset and choose authentication for the actual execution environment. A person still reviews the facts, rights, disclosures, destination, payload, and time. Submit once, capture the post ID, and verify accepted delivery. Reconcile an uncertain result before retrying.

Start from the execution boundary

An Amp setup can be correct locally and absent in an Orb, or overridden by a higher-precedence definition. Solve that boundary before working on the social payload.

Name the approved input first:

Canonical source asset:
Approved version:
Owner:
Facts and links verified by:
Rights and consent verified by:
Disclosure requirement:
Approved copy or adaptation limits:
Approved media:
Allowed destination and timing window:

Research, creation, recording, editing, approval, and rights or disclosure checks for this asset stay outside Groniz. Amp can assemble a tool call from the packet, while the account operator remains responsible for the public claims and final delivery decision.

For the protocol and publishing-layer distinction, see social media MCP servers. The client-by-client setup guide compares Amp’s precedence and Orb boundary with other confirmed MCP clients.

Map Amp’s configuration precedence

Amp’s official MCP documentation describes this precedence, from highest to lowest:

CLI flags
→ workspace configuration
→ user configuration
→ skills

The effective server comes from the highest applicable definition in that list, which may not be the one you edited most recently. If the same server name appears in more than one layer, inspect every applicable definition before the first write. Record which layer won so the runbook is unambiguous.

Use amp mcp add when the current documented remote-server flow fits the intended scope, or define the server under amp.mcpServers as the official guidance describes. A workspace MCP server requires approval. That approval establishes whether the workspace may offer the server; it does not approve a post.

Avoid putting a reusable credential in a workspace definition that may be committed or shared. A team can review a secret-free remote endpoint while each execution environment receives credentials through an appropriate protected mechanism.

Separate local Amp from Orbs

Local MCP settings do not automatically become available inside Orbs. Treat local and Orb execution as two deployment targets:

Question	Local Amp	Orb
Does local MCP config apply?	It may, according to the active precedence layer	No automatic inheritance
Can browser OAuth run in the execution environment?	Use it when the current local path supports it	Browser OAuth is unavailable inside Orbs
Where should the remote server be defined?	Effective CLI, workspace, user, or skill layer	Hosted remote-server settings for the Orb environment
How should a fallback secret be supplied?	Protected local mechanism	Secret-backed header appropriate to the hosted environment
What needs approval?	Exact write plus any applicable workspace server	Hosted server access, exact write, and destination packet

Do not "fix" an Orb connection by pasting a real token into the task prompt or repository. Configure the remote server and authentication in the hosted environment with the mechanisms Amp documents for that environment. Test it there instead of assuming the local result carries over.

The choice also affects file access. An Orb may not see a local approved asset merely because a local Amp session can. Provide the stable source through the intended environment and verify the version before preparing the payload.

Add the Groniz remote server

Use this endpoint:

https://mcp.groniz.com/mcp

Groniz remote MCP supports OAuth 2.0, an Authorization: Bearer YOUR_API_KEY header, or a key embedded in the URL. Prefer OAuth in a local environment when Amp’s current remote-server flow can complete it. Inside an Orb, browser OAuth is unavailable, so use hosted remote-server settings with authentication that works there, such as a secret-backed header, according to Amp’s current documentation.

If a fallback API key is required, the issuance location is Groniz Connectors API keys. Never put a real value in a command example, workspace file, prompt, screenshot, or delivery log. A key embedded in the endpoint URL is still a secret and is especially easy to expose in configuration views.

Run the connection test in the exact environment that will publish. Confirm the effective Groniz definition, the remote endpoint, authentication, and live tool discovery. Local success is not Orb evidence, and server discovery is not content approval.

Complete the precedence and boundary matrix

Use this original setup matrix for the first delivery:

Area	Record for this run	Required evidence
Approved source	Asset and immutable version	Owner and review reference
Execution	Local Amp or named Orb	Environment that will invoke the write
Winning config	CLI, workspace, user, or skill	Applicable definitions checked in precedence order
Workspace server	Applicable or not	Workspace approval when required
Remote endpoint	Groniz MCP URL	No real key in the recorded value
Authentication	OAuth or environment-appropriate secret-backed method	Method and credential owner, never secret value
Destination	Exact connected account, Page, channel, or community	Stable integration reference and readable label
Content gate	Exact payload, media, settings, and time	Reviewer and approval reference
Delivery proof	Accepted result	Post or scheduled-record ID and verification state

Complete every row before allowing a write. If the effective server or execution context changes, revisit authentication, available tools, source access, and approval. Do not carry the previous answers forward unchanged.

Inspect live tools and integration settings

Once the remote server is active in the correct environment, use read operations first:

discover live MCP tools and schemas
→ list connected integrations
→ choose the exact destination
→ inspect its current required settings
→ construct one destination-specific payload

Use the provider fields reported by the live Groniz integration instead of inventing them or copying them from another network. Groniz handles provider OAuth, per-platform formatting, and delivery to 32+ networks, but providers differ in their fields, media, analytics, and scheduling. The live integration settings are authoritative.

If multiple accounts have similar names, show a person the readable account label and stable reference. Do not let the repository name, Orb name, or current workspace choose the public destination implicitly.

Media must be uploaded before posting. Confirm the supported upload completed, record the returned reference with the destination packet, and preserve attachment order. A local file path is not a delivery reference, and an Orb may have different access to that path in any case.

Keep the final write approval explicit

Before invoking the publishing tool, have Amp display this complete proposal without credentials:

Execution environment: local / named Orb
Effective MCP configuration layer:
Source asset and version:
Network and exact connected account:
Final text, links, and disclosure:
Uploaded media references and order:
Live integration settings:
Publish now or schedule:
Local time, named timezone, and exact offset timestamp:
Write operation:

The reviewer checks the facts, rights, disclosure, destination, payload, media, settings, and timing together. Workspace-server approval does not cover the content. If Amp changes the payload, destination, or time after review, cancel the write and obtain a new decision.

The human approval guide can turn this gate into a repeatable policy. Ask the reviewer directly: may this exact packet go through this integration at this time?

Verify in the same environment that submitted

Submit the approved write once. Record the environment, winning configuration layer, integration, payload version, submission time, response state, and returned post or scheduled-record ID.

For immediate delivery, inspect the supported state and public destination when available. For a schedule, confirm the stored account, content, timestamp, timezone, and offset. Acceptance or scheduling is evidence of a workflow state, not automatic proof of a public post.

If a local session or Orb loses the response, keep the outcome unknown. Search the available records and intended destination for the original submission before switching environments or repeating the call. The failed-post recovery workflow provides the reconciliation sequence. Retry only when non-delivery is established and re-approve any corrected payload or time.

Store the run record with the environment that submitted the write. It documents a reviewed path, while reach, engagement, leads, sales, and revenue remain outside its scope.

Once the approved source and local-or-Orb boundary matrix are complete, connect Amp’s reviewed delivery path through Groniz Connectors.