Connect AI Assistants to Prospectory with the MCP Server

Prospectory now has a hosted Model Context Protocol server at https://api.prospectory.ai/api/v1/mcp. It lets a compatible AI assistant discover and call Prospectory tools for Lead Packs, Account Intelligence, and AWS co-sell work without a custom integration for every assistant.

The practical result is simple: you can ask an assistant to summarize your AWS co-sell position, find the most ready opportunities, review expansion recommendations, or retrieve account research. The assistant calls Prospectory's live tools, receives structured results, and explains them in the context of your question.

MCP is an open protocol for connecting AI applications to external tools and data ^[1]. Prospectory implements the Streamable HTTP transport defined by the protocol ^[2], exposes tool contracts through tools/list ^[5], and returns both human-readable text and structured content from tool calls.

The figures below come from the production MCP endpoint itself rather than from any outside source, so they describe this release and will move as the server does:

What the Prospectory MCP connection can do

The first release focuses on bounded tasks that are useful in account and partner workflows. It does not attempt to turn every product action into an agent command.

The AWS opportunity search accepts an account-name fragment, an exact status, a minimum readiness score from 0 to 100, and a result limit. Results are ordered by readiness, then by the latest update. The response also tells the caller that its source view is bounded to the 100 most recently updated local opportunities and up to four Partner Central pages. That disclosure matters because a useful ranked list is not the same as a complete pipeline export.

The MCP server also publishes safety annotations. Read-only tools declare that they do not modify data. The Account Intelligence deletion tool declares that it is destructive and non-idempotent, and its description instructs the client to obtain explicit confirmation.

Create a workspace API key

An administrator creates the credential in the Prospectory application:

1. 1Sign in at app.prospectory.ai.
2. 2Open Settings, then API Keys.
3. 3Select Create API Key.
4. 4Give the key a purpose-specific name, such as MCP for RevOps.
5. 5Copy the value when it appears. Prospectory displays the plaintext value only at creation time.

Store the key in a password manager or secret store. Do not paste it into a prompt, commit it to a repository, or put it in a shared client configuration file. Create separate keys for separate environments or integrations so each connection can be revoked without interrupting the others.

The key is sent in the x-api-key HTTP header. Prospectory validates the key against its stored hash, checks its status and expiration, and derives the workspace and user scope before any MCP method runs. MCP's authorization specification describes OAuth-based flows for HTTP transports ^[4], but this Prospectory release uses API-key authentication. That distinction affects which clients can connect directly.

Verify the connection before adding an AI client

You can test the handshake with any HTTP client. Read the credential through a hidden prompt, then export the resulting shell variable. The key value will not be recorded in shell history or displayed on screen.

read -rsp "Prospectory API key: " PROSPECTORY_API_KEY && echo
export PROSPECTORY_API_KEY

curl -sS https://api.prospectory.ai/api/v1/mcp \
  -X POST \
  -H "content-type: application/json" \
  -H "accept: application/json, text/event-stream" \
  -H "mcp-protocol-version: 2025-06-18" \
  -H "x-api-key: $PROSPECTORY_API_KEY" \
  --data '{
    "jsonrpc": "2.0",
    "id": 1,
    "method": "initialize",
    "params": {
      "protocolVersion": "2025-06-18",
      "capabilities": {},
      "clientInfo": {"name": "connection-check", "version": "1.0.0"}
    }
  }'

A successful response has HTTP status 200, identifies the server as prospectory-mcp, and returns protocol version 2025-06-18. Complete the MCP initialization lifecycle before listing tools:

curl -sS https://api.prospectory.ai/api/v1/mcp \
  -X POST \
  -H "content-type: application/json" \
  -H "accept: application/json, text/event-stream" \
  -H "mcp-protocol-version: 2025-06-18" \
  -H "x-api-key: $PROSPECTORY_API_KEY" \
  --data '{"jsonrpc":"2.0","method":"notifications/initialized"}'

curl -sS https://api.prospectory.ai/api/v1/mcp \
  -X POST \
  -H "content-type: application/json" \
  -H "accept: application/json, text/event-stream" \
  -H "mcp-protocol-version: 2025-06-18" \
  -H "x-api-key: $PROSPECTORY_API_KEY" \
  --data '{"jsonrpc":"2.0","id":2,"method":"tools/list","params":{}}'

The response should contain the 11 tools described above. Prospectory's current endpoint does not issue an Mcp-Session-Id because it handles these requests without server-side sessions. If a future server response includes that header, save it from the initialize response and send the same value on notifications/initialized and every later request in that session, as required by the Streamable HTTP transport ^[2]. A conforming MCP client handles this lifecycle automatically.

This direct check isolates credential and network problems from client configuration. A 401 response usually means the key is absent, invalid, revoked, or expired. A successful handshake followed by a client error usually points to the client's transport or header configuration.

Configure an MCP client

Use an MCP client that supports a remote Streamable HTTP server and custom request headers. Configuration shapes differ, but the essential values remain the same:

{
  "mcpServers": {
    "prospectory": {
      "type": "http",
      "url": "https://api.prospectory.ai/api/v1/mcp",
      "headers": {
        "x-api-key": "${PROSPECTORY_API_KEY}"
      }
    }
  }
}

Treat this as a field map, not a file to copy blindly. Your client may call the server field servers, mcpServers, or connections. It may also use a secure credential prompt instead of environment-variable expansion. Follow the client's current documentation and verify that it sends custom headers to remote MCP servers.

Some hosted assistants accept only OAuth-authenticated remote MCP servers. An x-api-key field may not be available in their connector UI. For those clients, do not place the key in the URL or prompt. Use a controlled authentication bridge that adds the header server-side, or wait for Prospectory's native OAuth option.

After saving the connection, ask the client to enumerate available Prospectory tools. Confirm the server name, tool names, input schemas, and read-only or destructive annotations before allowing execution.

Run the first three prompts

Start with questions whose results are easy to verify in the Prospectory application.

1. Workspace and pipeline briefing

Use Prospectory to summarize our AWS co-sell position. Include CRM connection health, AWS connection readiness, pipeline metrics, workflow status, and any guardrails. Do not modify data.

This should select prospectory_get_aws_cosell_summary. The structured result includes workspace context, CRM providers, AWS connection state, metrics, queue state, workflow state, and guardrails.

2. Prioritized opportunity review

Use Prospectory to find up to 10 AWS co-sell opportunities with readiness of at least 70. Rank them by readiness and explain the next review step. Treat the result as a bounded view, not a complete export.

This should call prospectory_find_aws_cosell_opportunities with minimumReadiness set to 70 and limit set to 10. Check that the assistant preserves the completeness note rather than presenting the output as the entire opportunity universe.

3. Expansion planning

Use Prospectory to retrieve evidence-backed AWS Marketplace expansion recommendations. Group them by recommended action and cite the evidence returned by the tool. Do not create tasks or update CRM records.

This uses prospectory_get_aws_expansion_recommendations. The response contains recommendations, generation time, status, a user-facing message, and the applicable data boundary.

Once those results match the AWS Marketplace area in Prospectory, test Lead Pack and Account Intelligence retrieval. Add write actions later, with narrow permissions in the surrounding workflow and a clear approval step.

A safe operating checklist

Review the key's Last Used value in Prospectory after testing. If an integration is retired or a key appears in a chat, ticket, log, or commit, revoke it and create a replacement. Do not depend on deleting a local configuration alone.

Frequently asked questions

What is the Prospectory MCP endpoint?

The production endpoint is https://api.prospectory.ai/api/v1/mcp. It uses the Streamable HTTP JSON transport and expects the API key in the x-api-key header.

Which Prospectory data can an assistant access?

The current tools cover Lead Packs, Account Intelligence, and AWS co-sell intelligence. Every request is scoped to the workspace associated with the API key. The assistant does not receive access to another Prospectory tenant through these tools.

Does it work with ChatGPT or Claude?

It works directly with MCP clients that support remote Streamable HTTP servers and custom headers. Some hosted ChatGPT or Claude connector experiences may require OAuth instead of an API-key header. Check the client's current connection requirements before configuring it, and use an approved authentication bridge if OAuth is mandatory.

Can the MCP connection change Prospectory data?

Yes, some tools can create a Lead Pack or create, sync, and delete Account Intelligence records. The three AWS co-sell tools are read-only. Configure your client to require explicit approval for write and destructive tools.

How should I troubleshoot a 401 response?

Confirm that the header is named exactly x-api-key, that the key is active and unexpired, and that you copied its full value. Create a replacement if the original plaintext value was lost or exposed.

Start with one repeatable workflow

The best first use is not an open-ended request to "analyze everything." Pick one recurring decision, such as the weekly AWS co-sell review. Define the filters, the expected evidence, the owner, and what the assistant must never change. Then compare its result with the Prospectory application for several cycles.

When that workflow is reliable, add the next one. The MCP connection supplies live, typed access to Prospectory. Your operating rules still determine whether an assistant's answer becomes a useful decision or an unreviewed action.

Create your key in Prospectory API Keys, connect the production endpoint, and begin with the three read-only AWS tools.

References

[1]Model Context Protocol, Introduction. https://modelcontextprotocol.io/introduction

[2]Model Context Protocol specification, Transports. https://modelcontextprotocol.io/specification/2025-06-18/basic/transports

[3]Model Context Protocol specification, Lifecycle. https://modelcontextprotocol.io/specification/2025-06-18/basic/lifecycle

[4]Model Context Protocol specification, Authorization. https://modelcontextprotocol.io/specification/2025-06-18/basic/authorization

[5]Model Context Protocol specification, Tools. https://modelcontextprotocol.io/specification/2025-06-18/server/tools