The Harvestr MCP server lets AI assistants like Claude and Cursor query your Harvestr workspace — your feedback, discoveries, customers, and components — directly in natural language.

There are two ways to connect a client:

OAuth (recommended for most users): a one-click browser flow inside your AI client. No tokens to copy or paste.
Personal access token: for headless setups, agentic workflows, or clients that don't support OAuth yet.

Server URL

https://mcp.harvestr.io/mcp

Connect via OAuth

OAuth is the default flow for most modern MCP clients. The process is the same regardless of the client: add the Harvestr MCP server in your client's settings, the client opens a browser, you authorize from inside Harvestr, and the client is connected.

Claude

In Claude, open Customize → Connectors → Add custom connector.
Enter the server URL: https://mcp.harvestr.io/mcp.
Save. A browser window opens prompting you to authorize the connection.
Sign in to Harvestr if needed, choose Read or Write scope, and confirm.
The connection appears in Settings → Integrations → MCP Server → Connected apps inside Harvestr. Harvestr tools become available in the client.

Cursor, Windsurf, and other OAuth-capable clients

The flow is the same. Open your client's MCP settings, add a new server with URL https://mcp.harvestr.io/mcp, then complete the OAuth prompt in your browser.

Scopes

When you authorize, you pick the level of access the client gets:

Read: browse and analyze your workspace data (feedback, discoveries, components, customers, activities).
Write: create, update, and delete discoveries and components, and move feedback between discoveries — orchestrated through plans (see Write tools below).

You can view every connected app, see which scope was granted, and revoke any connection at any time from Settings → Integrations → MCP Server.

Connect with a personal access token

Use a personal access token when:

Your AI client doesn't support OAuth yet.
You're running an agent or script in a headless environment with no browser to complete the OAuth flow (CI, server-side automations, n8n flows, custom orchestrators).

A token grants the client the same access as the user who created it.

Step 1: Generate a token

Go to Settings → Integrations → MCP Server.
Click New MCP Token.
Add a label to identify the client.
Click Create, then copy and store the token securely. You won't be able to see it again after this screen.

Each token is scoped to your organization and can be revoked at any time from the same page.

Step 2: Connect your AI client

In your MCP client, add a new server with:

URL: https://mcp.harvestr.io/mcp
Authentication: Bearer token → paste your MCP token

Save and reconnect.

What you can do

The Harvestr MCP server gives connected AI tools access to the following data and actions.

Read tools (available with both scopes)

Tool	What it does
`list_harvestr_discoveries`	Search and filter discoveries by title, component, state, or date. Primary tool for browsing your backlog.
`list_harvestr_discoverystates`	List all workflow states (Backlog, In Progress, Done, and so on) for filtering discoveries.
`list_harvestr_discovery_fields`	List custom and default discovery fields available in your workspace.
`list_harvestr_discovery_tags`	List the tags available on discoveries.
`list_harvestr_discovery_templates`	List the discovery templates configured in your workspace.
`list_harvestr_feedback`	Semantic search across feedback by topic, customer, or discovery. Returns both validated and AI-suggested feedback.
`list_harvestr_messages`	Access full source messages (support tickets, Slack threads, notes). Use when you need the complete context behind a feedback extract.
`list_harvestr_components`	Browse your product component hierarchy. Useful for scoping queries to a specific product area.
`list_harvestr_custom_inboxes`	List the Custom Inboxes set up in your workspace.
`list_harvestr_customers`	Look up users and companies by name. Useful for finding customer IDs before filtering feedback by requester.
`search_harvestr_customers_by_attributes`	Find customers by CRM attributes — plan tier, industry, geography, and so on. Supports semantic search on text attributes and numeric filters.
`analyze_harvestr_feedback`	Analyze feedback volume over time. Returns a time series with period-over-period changes, globally or broken down by discovery.
`read_harvestr_activities`	Read the activity log on a discovery or component for change history.
`read_harvestr_deleted_archive`	Browse archived and deleted items for recovery or audit.
`read_harvestr_plan`	Read the contents of a plan that's already been created.

Write tools (Write scope only)

Write operations go through Harvestr plans. Three tools handle plan orchestration:

Tool	What it does
`create_harvestr_plan`	Start a new plan that will package one or more actions to apply to your workspace.
`add_harvestr_plan_step`	Append an action to an existing plan.
`execute_harvestr_plan`	Run all the actions in a saved plan against your workspace.

The actions a plan can perform:

Action	What it does
`DISCOVERY_CREATE`	Create a new discovery.
`DISCOVERY_UPDATE`	Update an existing discovery.
`DISCOVERY_DELETE`	Delete a discovery.
`CHUNK_UPDATE`	Move a single feedback between discoveries.
`COMPONENT_CREATE`	Create a new component.
`COMPONENT_UPDATE`	Update an existing component.
`COMPONENT_DELETE`	Delete a component.

A few things worth knowing when building workflows

list_harvestr_feedback is the only tool with true semantic search — it finds conceptually related feedback even with different wording. Other list tools use keyword/title matching.
Tools chain naturally: use list_harvestr_discoveries to get a discovery ID, then list_harvestr_feedback with discovery_ids to get its feedback, then list_harvestr_customers to profile who's behind it.
All tools support pagination via a cursor parameter for workflows that need to process large result sets.

Security

Treat MCP tokens like passwords — never commit them to version control.
Each token and each OAuth connection is scoped to your organization only.
Use one token per client so you can revoke individual access without disrupting other tools.
Revoke tokens or OAuth connections anytime from Settings → Integrations → MCP Server.
For Write connections, audit the connected app's activity in your MCP Server settings regularly.

Troubleshooting

OAuth browser window doesn't open

Make sure your AI client is up to date. Older versions of some clients don't yet implement the MCP OAuth flow — try a personal access token instead.

"No tools available" after connecting

For token-based connections, check that your token is correctly formatted with the Bearer prefix in the Authorization header. For OAuth, try disconnecting from your client and re-authorizing.

Authentication errors

Your token or OAuth connection may have been revoked. Generate a new token, or re-authorize from your client. Check your active connections at Settings → Integrations → MCP Server.

Slow responses on large queries

Queries that span your full workspace (no date range or filter) may take longer. Scope your question to a time period or specific component for faster results.

Connect your AI tools to Harvestr with MCP