Power your agent
The publisher's own command. Not a PlugMCP endpoint.
Host a copy with our keys and logs.
npx -y @salesforge/[email protected]Claude Code has no install link for a local command. Copy the command below and run it in a terminal.
SALESFORGE_API_KEY=<secret> PRIMEFORGE_API_KEY=<secret> LEADSFORGE_API_KEY=<secret> INFRAFORGE_API_KEY=<secret> MAILFORGE_API_KEY=<secret> claude mcp add io.github.salesforgeai-forge-mcp -- npx -y @salesforge/forge-mcp@1.0.0Run in a terminal.
{ "mcpServers": { "io.github.salesforgeai-forge-mcp": { "command": "npx", "args": [ "-y", "@salesforge/[email protected]" ], "env": { "SALESFORGE_API_KEY": "<secret>", "PRIMEFORGE_API_KEY": "<secret>", "LEADSFORGE_API_KEY": "<secret>", "INFRAFORGE_API_KEY": "<secret>", "MAILFORGE_API_KEY": "<secret>" } } }}claude_desktop_config.json
Remote URL or a command on your machine? How the two installs differ
Run on PlugMCP
A new server you create. It is not this listing's official endpoint. Use the publisher install when you want their URL and their credentials.
LaunchFrom the publisher
A remote MCP server that connects AI assistants to the full Salesforge product suite: Salesforge, Primeforge, Leadsforge, Infraforge, Warmforge, and Mailforge.
Built on the Model Context Protocol, works with Claude Desktop, Claude Code, Cursor, Windsurf, and any MCP-compatible client.
MCP server for Salesforge, Primeforge, Leadsforge, Infraforge, Warmforge, and Mailforge
Salesforge (78 tools) - Account credits and subscription, workspaces, contacts, sequences and subsequences, lead progress, mailboxes, LinkedIn accounts, sender profiles, tags, enrollment preflight and confirmation, webhooks, email validation, do-not-contact lists
Primeforge (23 tools) - Workspaces, domains, mailboxes, DNS management, prewarmed mailboxes, bulk mailbox lookup
Leadsforge (41 tools) - Contact search and count, email/phone/LinkedIn enrichment, company followers, lookalike search and free preview, local business discovery and owner enrichment, filter value lookups
Infraforge (24 tools) - Workspaces, domains, mailboxes, DNS, domain availability, credits
Warmforge (15 tools) - Workspaces, mailboxes, warmup stats, placement tests, latest mailbox placement results
Mailforge (23 tools) - Workspaces, domains, mailboxes, DNS management, domain availability, auto-renewal, domain masking, forwarding
Only provide API keys for the products you use. Tools for unconfigured products won't appear.
Add to your claude_desktop_config.json:
{
"mcpServers": {
"salesforge": {
"url": "https://mcp.salesforge.ai/mcp",
"headers": {
"X-Salesforge-Key": "YOUR_SALESFORGE_API_KEY",
"X-Primeforge-Key": "YOUR_PRIMEFORGE_API_KEY",
"X-Leadsforge-Key": "YOUR_LEADSFORGE_API_KEY",
"X-Infraforge-Key": "YOUR_INFRAFORGE_API_KEY",
"X-Warmforge-Key": "YOUR_WARMFORGE_API_KEY",
"X-Mailforge-Key": "YOUR_MAILFORGE_API_KEY"
}
}
}
}
Remove header lines for products you don't use. Restart Claude Desktop after saving.
claude mcp add salesforge \
--transport streamable-http \
--url https://mcp.salesforge.ai/mcp \
--header "X-Salesforge-Key: YOUR_SALESFORGE_API_KEY" \
--header "X-Primeforge-Key: YOUR_PRIMEFORGE_API_KEY" \
--header "X-Leadsforge-Key: YOUR_LEADSFORGE_API_KEY" \
--header "X-Infraforge-Key: YOUR_INFRAFORGE_API_KEY" \
--header "X-Warmforge-Key: YOUR_WARMFORGE_API_KEY" \
--header "X-Mailforge-Key: YOUR_MAILFORGE_API_KEY"
Go to Settings > MCP and add a new server:
If you manage multiple accounts (for example, different clients), add separate server entries:
{
"mcpServers": {
"salesforge-client-a": {
"url": "https://mcp.salesforge.ai/mcp",
"headers": {
"Authorization": "Bearer CLIENT_A_KEY"
}
},
"salesforge-client-b": {
"url": "https://mcp.salesforge.ai/mcp",
"headers": {
"Authorization": "Bearer CLIENT_B_KEY"
}
}
}
}
Each entry gets its own name and API keys. Your AI assistant sees tools from both and you specify which client to work with in your prompts.
After setup, try asking your AI assistant:
Start validation with start_email_validation. Before returning HTTP 201, Multichannel resolves and persists candidate membership. selected and skipped.duplicate are exact, supported counts: a non-empty selection returns status: pending, while an empty selection returns status: failed with failureCode: validation_scope_empty. The legacy strict option is omitted and deprecated. Retain the returned validationJobID and poll get_validation_results with roughly ten-second backoff. Continue until the status is terminal: completed, partially_completed, or failed. Inspect the polling response before enrolling contacts; it is authoritative for validation outcomes, terminal counts are authoritative, and a failed run must not be automatically resubmitted.
For enrollment, use preflight_enrollments with the completed or partially completed validationRunId, inspect the preflight, then use confirm_enrollment_preflight. To restrict candidates, provide non-empty validationStatuses; omitted or empty statuses mean no status restriction and all run candidates, including unvalidated contacts, are considered. Durable async runs evaluate non-empty statuses against the requested run's persisted results, while legacy completed runs evaluate them against current Salesforge status. A partial run does not automatically select only its successful subset.
Local HTTP MCP testing still proxies the deployed Multichannel API.
Internal rollout order is binding: deploy the Multichannel async-validation contract before rolling out Forge MCP. Forge MCP must not reintroduce strict: false as a rollout workaround.
src/
├── index.ts # stdio entry point
├── http.ts # HTTP/SSE transport entry point
├── server.ts # MCP server setup, tool registration
├── api-client.ts # HTTP client for upstream APIs
├── client.ts # legacy client
├── helpers.ts # shared utilities
└── tools/
├── identity.ts # API key validation
├── account.ts # account credit balances and subscription
├── workspaces.ts # workspace management
├── contacts.ts # contact CRUD
├── mailboxes.ts # mailbox and email operations
├── threads.ts # primebox threads, labels, and LinkedIn replies
├── sequences.ts # sequence lifecycle and lead progress
├── subsequences.ts # subsequence triggers, parents, and members
├── nodes.ts # sequence node management
├── branches.ts # sequence branches
├── enrollments.ts # contact enrollment
├── linkedin.ts # LinkedIn connection and session management
├── tags.ts # workspace tag discovery
├── sender-profiles.ts# sender profile management
├── validations.ts # email validation
├── webhooks.ts # webhook management
├── dnc.ts # do-not-contact lists
├── custom-vars.ts # custom variables
├── reference.ts # action/condition type lookups
├── primeforge/ # domain, mailbox, workspace tools
├── leadsforge/ # search, enrichment, lookalike, followers, local business tools
├── infraforge/ # domain, mailbox, credit tools
├── warmforge/ # workspace, mailbox, placement test tools
└── mailforge/ # workspace, domain, mailbox tools
If you prefer to run your own instance:
npm install
npm run build
npm run start:http
The server listens on port 3000 by default. API keys are passed as headers per request, not as environment variables.
The HTTP endpoint uses stateless MCP POST requests. Standalone SSE streams are not offered: GET /mcp (and HEAD) returns 405 Method Not Allowed before allocating an MCP server. Per-request servers and transports are closed after handling, on failures, and when clients disconnect before the response finishes.
Logging is enabled by default as JSON lines on stderr, suitable for container log collection. MCP stdio output remains reserved for protocol messages. HTTP requests, including /health, receive a generated X-Request-Id response header that correlates their request and upstream API logs.
For more detail, run LOG_LEVEL=debug npm run start:http. Invalid settings fall back to their defaults. Keep info or debug enabled during timeout investigations so request starts and process metrics are retained.
http.request.started without a matching completed or aborted event indicates an unfinished request or a process that stopped before logging completion. http.request.slow is emitted while the request is still pending, including during body parsing.upstream.request.slow identifies a pending product API call; its timing includes reading and parsing the response. The parent requestId links it to the HTTP request, and upstreamRequestId separates multiple API calls.runtime.metrics shows whether event-loop delay, CPU, memory, or request counts increased around a timeout. A blocked event loop also delays logging; its stall becomes visible after it resumes.The structured logs omit headers, bodies, query strings, upstream paths, and error messages/stacks to avoid capturing API keys and customer data. Errors include type, code, and HTTP status where available. Existing Node/Express native diagnostics may still appear separately. HTTP routes other than /health and /mcp are labeled other.
MIT
Use get_thread to read the conversation, then call reply_to_linkedin_thread with workspaceId, threadId, the sending LinkedIn accountId from the thread's LinkedIn messages, and message. LinkedIn-only threads do not require a mailbox ID. The tool calls the existing Salesforge public API and returns the created LinkedIn message.
Optional attachments accept filename, contentBase64, and optional contentType. For an attachment-only reply, pass an empty message and at least one attachment. The API validates supported file types and size limits.
These tools require the public API additions in multichannel-api #1068 and salesforge-api #2795 (SF-9461). The matching API changes must be deployed before the tools can be used.
To attach a standalone LinkedIn account, use create_sender_profile with its linkedinAccountId. The account must belong to the workspace and must not already be attached to another sender profile.
The existing delete_sender_profile tool also deletes its attached LinkedIn account.
These tools require salesforge-api #2836 and multichannel-api #1111 (SF-10349), on top of the SF-9461 API additions above. The matching API changes must be deployed before the tools can be used.
There is no independent LinkedIn deletion tool or email disconnect tool. The existing delete_sender_profile tool also deletes its attached LinkedIn account.
© 2026 MCP Directory.
MCP server for Salesforge, Primeforge, Leadsforge, Infraforge, Warmforge, and Mailforge
{ "mcpServers": { "io.github.salesforgeai-forge-mcp": { "command": "npx", "args": [ "-y", "@salesforge/[email protected]" ], "env": { "SALESFORGE_API_KEY": "<secret>", "PRIMEFORGE_API_KEY": "<secret>", "LEADSFORGE_API_KEY": "<secret>", "INFRAFORGE_API_KEY": "<secret>", "MAILFORGE_API_KEY": "<secret>" } } }}Tools
No tool list yet. Remote servers get one after the first health check.
FAQ
Copy the stdio config for Cursor, Claude Desktop, Claude Code, or VS Code on this page. It uses the publisher's install, not a PlugMCP proxy.
The publisher lists these environment variables: SALESFORGE_API_KEY, PRIMEFORGE_API_KEY, LEADSFORGE_API_KEY, INFRAFORGE_API_KEY, MAILFORGE_API_KEY.
Forge runs locally over stdio.
Last checked 2026-11-09. MCP server for Salesforge, Primeforge, Leadsforge, Infraforge, Warmforge, and Mailforge