Keywords AI MCP server: 55 tools, now Respan

The product changed name, and the connector is large. What its nine categories cover, and the header that registers only the tools you name.

Last verified 27 August 2026 · from The vendor's own MCP documentation, published under the Respan name

This page is one of 90 in a directory of monitoring and developer tools. Each page checks what an AI coding agent can get out of the tool through its MCP server, the connector that lets the agent query the tool directly.

What is the Keywords AI MCP server?

It is Respan's server for the Model Context Protocol. The product traces language-model calls and evaluates them, so a connected agent reaches both the traffic and the scoring.

The reading surface is the useful half for most sessions. list_logs and get_log_detail cover request history, while list_traces, get_trace_tree and get_traces_summary cover structure.

The authoring half covers prompt versions, evaluators and workflows, which most observability connectors do not expose. Prompts have versions that can be created and deployed, and evaluators and workflows have their own commit tools.

It is strongest in agent observability, touches cost through logged usage, and reaches nothing in code review or CI.

For an agent, that means Respan answers what your model calls did, what they scored and what they cost. It can also deploy the next prompt version. It sees nothing outside model traffic, so a service question needs a separate connector.

What can an AI agent do with Keywords AI?

Decide the tool scope before the first session. With 55 tools registered by default, the Respan-Enabled-Tools header is the difference between a focused connection and a large one.

What connecting costs before the first question

A context window is the amount of text a model can hold at once. Fifty-five tools cost 2,815 to 3,407 tokens on Claude if the whole surface loads. Each description is one short line. That is why the tool count is high and the token cost is not.

The Respan-Enabled-Tools header narrows the set to named tools, so a scoped session pays less than this. Within the full 55, the figure is still a floor: no input schemas were available to count.

Encodings: GPT-5.6, GPT-5.5 o200k_base; GPT-4, GPT-3.5 cl100k_base; Claude Haiku 4.5 to Opus 5 count_tokens. Counted from the 55 tool entries in the vendor's own documentation. That is the only public inventory.

The server could not be run for a live tools/list. So no input schemas are available. These figures cover each tool's name and description only. They are a floor, not the figure.

The Claude columns are measured through the API's count_tokens endpoint, with an empty schema per tool. So they cover the same content as the GPT columns. The range there is two tokenizer generations. Haiku 4.5 and Opus 4.6 sit at the low end. Opus 5 and Sonnet 5 sit at the high end. The gap between them is about a fifth.

Can Keywords AI watch your own AI agents?

This is the most complete agent-observability surface measured in this guide, with only cost left partial. Read from the vendor's own documentation on 24 August 2026.

Keywords AI MCP server limits: where answers come back incomplete

Keywords AI starts with a large default surface and a changed brand; those two facts shape the limits below.

How to configure Keywords AI MCP for agents

1. 1

   Whitelist the tools you need

   Respan-Enabled-Tools takes a comma-separated list of exact tool names; without it, all 55 are available.

2. 2

   Prefer the API key

   The documentation recommends it, with OAuth and a JWT token as fallbacks.

3. 3

   Decide who may deploy

   deploy_prompt_version changes what production serves, and commit_workflow fixes a workflow version.

4. 4

   Know the name

   Everything is published as Respan, so searching for Keywords AI documentation will send you in circles.

Paste this into your agent’s instructions

Deployment tools sit in the same list as reading tools, and the tool scope can differ per connection. Both belong in the answer.

When you answer using Respan MCP tools, state:
- The project and period the answer covers.
- Whether the session was scoped with Respan-Enabled-Tools,
  and which tools were enabled.
- Whether you deployed anything. deploy_prompt_version
  changes what production serves.
Never state that data does not exist. State that no logs or
traces matched the filters above.
Write your answer in ASD-STE100 Simplified Technical
English. Use short sentences with one idea in each.

Do you need the MCP server at all?

Not for a scheduled export. Respan has a REST API, and for that job it avoids both the context cost and the write surface.

The connector earns its place when the loop is short: read the failing trace, adjust the prompt version, and evaluate it without leaving the editor.

For tracing alone, Langfuse and Arize Phoenix cover that ground with narrower connectors.

Connect a scoped session for prompt and evaluation work. Use the API for reporting.

Keywords AI MCP server setup

The server is hosted, so nothing is installed. An API key is the documented recommendation, passed as a bearer token.

OAuth is supported, with a JWT token documented as the fallback where an OAuth flow cannot complete.

claude mcp add --transport http respan \
  https://mcp.respan.ai/mcp \
  --header "Authorization: Bearer <your-respan-api-key>"

{"mcpServers":{"respan":{
  "url":"https://mcp.respan.ai/mcp",
  "headers":{"Authorization":"Bearer <your-respan-api-key>"}}}}

[mcp_servers.respan]
url = "https://mcp.respan.ai/mcp"
headers = { Authorization = "Bearer <your-respan-api-key>" }

Every other client

Each block below is the configuration for one client, with the file path and the key that client expects.

{"mcpServers":{"respan":{"url":"https://mcp.respan.ai/mcp",
  "headers":{"Authorization":"Bearer <your-respan-api-key>"}}}}

{"mcpServers":{"respan":{"url":"https://mcp.respan.ai/mcp",
  "headers":{"Authorization":"Bearer <your-respan-api-key>"}}}}

{"servers":{"respan":{"url":"https://mcp.respan.ai/mcp",
  "headers":{"Authorization":"Bearer <your-respan-api-key>"}}}}

{"mcpServers":{"respan":{"url":"https://mcp.respan.ai/mcp",
  "headers":{"Authorization":"Bearer <your-respan-api-key>"}}}}

claude mcp add --transport http respan \
  https://mcp.respan.ai/mcp \
  --header "Authorization: Bearer <your-respan-api-key>" \
  --header "Respan-Enabled-Tools: list_logs,get_log_detail"

Read-only access, permissions and security

• Prompts, prompt versions, evaluators, pipelines, workflows, datasets and logs can all be created or changed

  Three of the nine categories are write categories.

• One tool deploys

  deploy_prompt_version changes what production serves, from the same connection that reads logs.

• Deletion is reachable

  delete_evaluator, delete_dataset and delete_experiment all appear in the documented list. The tool groups above do not reproduce every one of the 55; delete_experiment is one they omit.

• Traces carry prompt content

  A trace tree includes the model calls it describes, so that content reaches the model reading it.

Troubleshooting

The endpoint returns 401

An unauthenticated request is rejected with an Unauthorized error. Pass the API key as a bearer token, or complete OAuth.

Documentation cannot be found under Keywords AI

The product is published as Respan. The MCP documentation lives under the Respan domain rather than the original brand.

All 55 tools appear in the client

All 55 tools register unless narrowed. Add the Respan-Enabled-Tools header with the exact tool names you want.

Production behavior changed after a session

Check for a deployment. deploy_prompt_version is in the default tool list.

Keywords AI MCP server: Reference

What engineers report

The documentation lists tools in per-category tables, so the counts below are its own.

Tool inventory, authentication routes and the scoping header read from the vendor's own MCP documentation on 24 August 2026, published under the Respan name. An unauthenticated tools/list against mcp.respan.ai/mcp returned 401 on the same date.

Should you connect an agent to Keywords AI?

Yes. Best for a scoped enabled-tools list per session, with deploy_prompt_version left out unless deploying is intended.

Strongest fit

Teams already tracing model calls with this platform who want prompt and evaluation work inside the editor.

Main advantage

A documented header for registering only the tools a session needs, which few connectors here provide.

Main weakness

The rename means most existing references point at a brand the documentation no longer uses.

Operational risk

Deployment and deletion tools register by default alongside the reading ones.

Frequently asked questions