MCP
Growhold works the same from its workspace and from the AI you already use. Through the Model Context Protocol, any MCP-capable assistant gets Growhold's live market intelligence, agents, builds and reviews as tools while you research, plan or write code.
Growhold in any AI
MCP is the open standard AI assistants use to call tools. Growhold publishes one MCP server, and every client that speaks the protocol can use it: Claude on the web and desktop, Claude Code, Cursor, Windsurf, VS Code assistants and your own agents. You connect once and then ask in plain words; nothing here is a developer feature.
What the connector gives your AI
Everything the workspace can do, as tools the assistant calls mid-conversation: search live buyer demand and what is already being built, read market gaps and buyer clusters, list your builds and their deliverables, start an agent in a build and read its cards, check whether a build's deliverables still agree, and add to the knowledge base. The results are the same cards, runs and reviews you see in the workspace, and they appear there too.
What to ask for
- “What are accelerator programme managers paying for right now, and where is supply thin?”
- “Give me the brief and the deliverables of the Verdict Evals build before I plan the sprint.”
- “Run the Gap Validator in that build on this idea, then read the verdict back to me.”
- “Do the deliverables in this build still agree? Show me what needs review.”
- In a coding assistant: “Pull the solution design and its business rules from Growhold and implement the first slice.”
What stays your decision
Reading is free-form. Changing a build is not: starting a run spends credits like it does in the workspace, and accepting or dismissing a deliverable review is only done when you say so. The server tells the assistant this, and the tools follow the permissions of the key or connector you created.
Connect
Two ways in. Both start on the API & Devs page of your workspace. The server address is https://api.growhold.com/mcp.
Sign in from the assistant (OAuth)
For assistants that add connectors by signing in, such as Claude on the web and desktop: create an OAuth client on the API & Devs page, then add Growhold as a custom connector in the assistant with the server address, the client id and the secret. The assistant signs in with your workspace and the tools appear in every conversation. Revoking the client disconnects it at once.
Use an API key
For assistants and agents configured by file or command, such as Claude Code, Cursor and Claude Desktop: create an API key on the API & Devs page and give the client the server address with the key as a bearer token. Keys are scoped by area, so a key with only the knowledge permission gives an assistant market intelligence and nothing else.
Client configuration
Every client takes the same two things: the address and the Authorization header. The shape most clients use:
{
"mcpServers": {
"growhold": {
"url": "https://api.growhold.com/mcp",
"headers": { "Authorization": "Bearer gh_your_api_key" }
}
}
}| Client | Where it goes |
|---|---|
| Claude (web, desktop, mobile) | Settings, Connectors, Add custom connector: use OAuth as above |
| Claude Desktop (key) | claude_desktop_config.json, the block above |
| Claude Code | claude mcp add --transport http growhold https://api.growhold.com/mcp --header "Authorization: Bearer gh_your_api_key" |
| Cursor | .cursor/mcp.json in the project, the block above |
| Other MCP clients | Streamable HTTP transport with the address and header above |
Tools
The server exposes the same actions as the REST API, as tools. Names follow the API action with an underscore: runs.start is runs_start.
Available tools
| Tool | Permission | Description |
|---|---|---|
knowledge_base_list | knowledge | List knowledge entries |
knowledge_base_add | knowledge | Add text or file as knowledge |
knowledge_base_remove | knowledge | Remove a knowledge entry |
knowledge_base_apply | knowledge | Apply knowledge to agent/specialist |
builds_list | runs | List builds: the products or ideas work is organised under |
builds_get | runs | One build with its recent runs and outcomes |
builds_create | runs | Create a build for a new product or idea |
runs_list | runs | List agent runs with their build |
runs_get | runs | Get run details with build and cards |
runs_start | runs | Start an async agent run in a build (build_id required, or none) |
runs_stop | runs | Stop a running agent |
runs_delete | runs | Delete a run and its cards (needs confirm) |
runs_pause | runs | Pause a running agent |
runs_resume | runs | Resume a paused agent |
runs_cards_list | runs | List output cards for a run |
runs_cards_get | runs | Get card with signed file URLs |
reviews_list | runs | Deliverable reviews in a build: what no longer agrees after a later run |
reviews_check | runs | Compare a run with the build's other deliverables |
reviews_accept | runs | Apply a proposed tune to the deliverable (user decision) |
reviews_dismiss | runs | Leave the deliverable as it is (user decision) |
knowledge_cards_list | knowledge | List knowledge cards |
knowledge_cards_get | knowledge | Get a knowledge card |
insights_overview | knowledge | Market overview: top demand categories, gaps and buyer clusters |
insights_clusters | knowledge | Demand clusters for a mode (build, automate, deliver) and vertical |
search_demand | knowledge | Semantic search over demand statements (what buyers need) |
explore_demand_clusters | knowledge | Browse demand clusters by size, growth or nearest to a query |
demand_cluster_detail | knowledge | Statements inside one demand cluster |
query_demand | knowledge | Count buyer requests by industry, buyer type, intent and more |
find_demand_tags | knowledge | Most frequent tools, skills and concepts across buyer requests |
search_supply | knowledge | Semantic search over solution statements (what is being built) |
explore_supply_clusters | knowledge | Browse solution clusters by size or nearest to a query |
supply_cluster_detail | knowledge | Statements inside one solution cluster |
query_supply | knowledge | Count products by industry, product type, workflow or seller type |
find_supply_tags | knowledge | Most frequent integrations, concepts and topics across products |
search_buyers | knowledge | Semantic search over buyer persona statements (who the buyers are) |
explore_buyer_clusters | knowledge | Browse buyer clusters (personas) by size or nearest to a query |
buyer_cluster_detail | knowledge | Persona statements in one buyer cluster, plus what they ask for |
schedules_list | schedules | List schedules |
schedules_create | schedules | Create a schedule |
schedules_update | schedules | Update a schedule |
schedules_delete | schedules | Delete a schedule |
schedules_pause | schedules | Pause a schedule |
schedules_resume | schedules | Resume a schedule |
agents_list | agents | List agents |
agents_get | agents | Get agent details |
specialists_list | specialists | List specialists |
specialists_get | specialists | Get specialist details |
Permissions and credits
A tool is offered only when the key or connector carries its permission: knowledge, runs, schedules, agents or specialists. Market intelligence queries, chat and agent runs spend the workspace's credits at the same prices as in the app; listing and reading cost nothing.
Advanced
For your own agents or a client without MCP support, the server speaks plain JSON-RPC 2.0 over HTTP.
Raw JSON-RPC
List tools
curl -X POST https://api.growhold.com/mcp \
-H "Authorization: Bearer gh_your_key" \
-H "Content-Type: application/json" \
-d '{"jsonrpc": "2.0", "id": 1, "method": "tools/list"}'Call a tool
curl -X POST https://api.growhold.com/mcp \
-H "Authorization: Bearer gh_your_key" \
-H "Content-Type: application/json" \
-d '{
"jsonrpc": "2.0",
"id": 2,
"method": "tools/call",
"params": {
"name": "search_demand",
"arguments": { "query": "monthly KPI collection from portfolio companies", "limit": 10 }
}
}'